1. ChatGPT iterates rapidly, OpenAI models evolve, and downstream applications bloom
GPT-4 is a large-scale multimodal model that adds image capabilities and more accurate language understanding than the previous generation of ChatGPT. Behind the upgrade of GPT is the further evolution of OpenAI's large language model, while driving the expansion of downstream applications, and emerging a new batch of application scenarios.
1.1. GPT iterative update, artificial intelligence set off a wave of technology
GPT has been upgraded to the fourth generation, and the model capability has been rapidly improved. ChatGPT is a natural language generative model developed by OpenAI, using the Transformer neural network architecture (also known as GPT-3.5 architecture), based on a large corpus using instruction learning and human feedback reinforcement learning (RLHF) to guide model training. Models that understand and generate human-like responses to a variety of topics are the result of advances in AIGC technology. On March 15, 2023, OpenAI officially announced the multimodal large model GPT-4, which expands the input content to text and images within 25,000 words, which can handle more complex and subtle problems than ChatGPT.
ChatGPT offers a transformative user experience with a skyrocketing number of users. ChatGPT exploded after its release, with more than one million users in just 5 days, and more than 100 million users 2 months after its launch, with visits increasing from 18.3 million to 672 million, making it the fastest growing consumer-grade app in history. The latest version of GPT-4 mainly improves the ability of language model on the basis of ChatGPT's GPT-3.5, and adds multi-modal functions, which can accept image input and understand image content, and the acceptable text input length has also increased to 32,000 tokens, and the performance of different language scenarios and internal adversarial authenticity assessment is significantly better than GPT-3.5, and has shown human level on various professional and academic benchmarks, providing users with a transformative user experience.
OpenAI models can accomplish the vast majority of text and image tasks with mature productivity. According to the official introduction of OpenAI, GPT-4 can accept text and image information and allow users to specify any image or language task, processing text of more than 25,000 words. In addition to outstanding performance in various standardized exams and different language situations, GPT-4 can directly read and analyze papers with pictures in image processing and analysis, undertake text, audio, image generation and editing tasks, and iterate creative and technical writing tasks with users, such as composing songs, writing scripts or learning users' writing styles.
ChatGPT has driven the AI trend, and a variety of related products have been launched. In the office field, Microsoft integrated GPT-4 into Office applications, and Chat GPT was integrated into Teams Premium to provide functions such as artificial intelligence generated chapters and real-time translation of subtitles; In the programming space, Viva Sales will leverage ChatGPT to generate response suggestions for customer questions in emails, and Stripe will use GPT-4 to scan business websites and provide summaries to customer support; In the field of software, Duolingo built GPT-4 into a new language learning subscription layer, and Baidu's "Wen Xin Yiyan" was officially launched, and the trend of AI began to spread across many industries at home and abroad.
1.2. AIGC has developed and reformed productivity and has the ability to apply it in multiple fields
AIGC has developed and reformed productivity and has multi-field application capabilities. AIGC can use a large amount of unlabeled data for self-supervised learning and then use a small amount of labeled data for transfer learning, which can continue to produce large-scale, high-quality, low-cost content, and has a revolutionary increase in productivity. In terms of application, AIGC has been relatively mature in text and code writing, image recognition and generation, and has been applied on a large scale in the fields of film and television, media, e-commerce, and C-end entertainment, and will gradually expand to other fields such as video and games in the future, providing more possibilities for innovation and development in various industries and fields.
For example, OpenAI + traditional search engine = AI intelligent search engine: ChatGPT can understand and generate human-like thinking and context-related results, while traditional search engines only give links to web pages, which need to be read, clicked and judged by users. Whether it is the efficiency of knowledge acquisition, the depth of answering questions, or the scalability of interaction, ChatGPT is a subversive evolution for traditional search engines.
OpenAI+ office software: Taking Office Family Bucket and Microsoft Teams premium ChatGPT as an example, ChatGPT+Office three major Word, Excel, and PowerPoint can complete the whole service from text creation to visualization. ChatGPT can automatically compose document content according to the user's description needs, shortening the document creation time; Intelligently recommend document formatting, typesetting, etc. according to the formed document content; And make personalized PPT according to user descriptions, and intelligently generate reading language to help PPT presentations. Teams Premium ChatGPT also has a variety of intelligent services: 1) automatically provide real-time translation; 2) Automatically generate text-based meeting minutes; 3) Automatically record the dialogue and content of the speaker; 4) Automatically generate tasks based on the discussion context, etc.
1.3.AI Landing in actual scenarios, the importance of edge computing power is accelerated
The trend of intelligent terminals accessing artificial intelligence large models is clear, and it is expected that AI will soon bloom at the downstream application level. The domestic semiconductor industry has accelerated the "innovation wheel", and fully prepared for key links such as computing, perception, and storage on the end side to meet new opportunities in artificial intelligence. As a major supplier of artificial intelligence computing power, NVIDIA has put forward the prediction of "embodied intelligence" for the next wave of artificial intelligence, that is, by combining intelligent algorithms with the perception, action and environmental interaction of physical entities, so that machines can interact with the environment and solve problems in a more natural and intelligent way, opening up the imagination space for further integration of AI and machines. In addition, the recent expansion of the number of open source plug-ins by Open AI is expected to quickly open the application of GPT in vertical fields.
From the perspective of the AI industry chain, the semiconductor and communication industry is the underlying foundation, electronic hardware is the terminal carrier, and media and computer focus on landing applications. The AI chip industry at the core of the infrastructure layer is a high-tech barrier industry, which is currently mainly monopolized by NVIDIA, and the gap between the mainland and overseas markets is obvious, with outstanding shortcomings and time cost investment; There are certain opportunities at the algorithm layer, but because the training of the model is a long payback period and high capital investment work, the final market will show the Matthew effect, which will be evolved by a few large manufacturers with general large model capabilities and towards infrastructure. As a terminal carrier carrying applications, electronic hardware will usher in an increase in demand driven by the development of various innovative applications.
In the process of AI commercialization, the application of "end" with the blessing of edge computing is expected to make up for the limitations of AI large models. Edge computing refers to moving computing and data storage capabilities to edge devices close to the data source, such as AI edge computing boxes, IoT devices, etc., rather than relying solely on remote cloud servers. Edge computing has natural advantages in cost, latency, and privacy, and can also be used as a bridge to preprocess massive complex requirements and direct them to large models. The future of AI requires both device-side and cloud-based AI. Running AI applications on the endpoint side improves cost-effectiveness, enhances privacy, personalization, and reduces latency; Compared to AI workload processing only on CPUs or GPUs, the Snapdragon platform integrates a dedicated Qualcomm AI engine to process AI workloads more efficiently, allowing small and thin devices to support device-side AI. According to STL Partners, the global edge addressable market will reach $445 billion in 2030, up from 48% in 10 years. According to Yiou Think Tank, the mainland edge computing market space will reach 198.8 billion yuan in 2025, which has great development potential.
2. C-end: AI+ broadens application scenarios and promotes innovation demand
With the release of lightweight large models, AI applications are expected to extend from PCs to mobile phones, IOTs, etc., enabling all kinds of smart terminals.
2.1. XR: AIGC reduces costs and increases efficiency to expand application scenarios, which is expected to increase device penetration
2.1.1. Terminal equipment brand manufacturers vigorously promote the expansion of application scenarios
Apple released the first MR headset Apple Vision Pro as scheduled, with full specifications at the hardware level, M2 computing + R1 sensing dual chip, equipped with a monocular 4k resolution silicon-based OLED+3p pancake display solution, and through 12 cameras + 5 sensors + 6 microphones, immersive acoustic system, two-way perspective, iris recognition, interpupillary distance adjustment, etc., only rely on voice, vision, feel and so on can be efficiently interacted in real time. Vision Pro focuses on gaming, work, and communication, and is expected to become a tool to enhance productivity in many fields such as industry, healthcare, and design, and also means that it has officially entered the era of spatial computing. Vision OS offers several use cases:
⚫ Work efficiently and break the geographical restrictions of remote work. Vision OS allows applications to be displayed side-by-side at any scale and provides unlimited screen real estate. Vision Pro is compatible with and displays Mac screens in virtual space and syncs with iPhones, iPads, and Macs. FaceTime makes group communication easier and more natural, making office scenes more comfortable and efficient.
⚫ 3D cameras add a new dimension to photos and videos. Spatial calculations make photos and videos stunning in color and detail, and panoramic photos can be unfolded and surrounded by life-size for an immersive feeling. Vision Pro is Apple's first 3D camera that leverages 3D capabilities and spatial audio to capture and store cherished memories.
⚫ Spatial transformation for a great audiovisual experience. With two 23-megapixel Micro OLED displays, Apple Vision Pro transforms any space into a personal theater with a screen as wide as 100 feet, and Apple Immersive Videos delivers 180-degree high-resolution video with spatial audio.
⚫ Vision Pro delivers immersive gaming. Spatial computing makes new game genres a reality, offering a variety of immersive levels of experience and support for commonly used gamepads. Apple partnered with Unity to introduce popular Unity games and apps that provide full access to Vision OS features such as perspective, high-resolution rendering, and native gestures, as well as simultaneous rendering with other Vision OS apps, creating a great platform for game developers.
Vision Pro opens up endless possibilities for developers. Developers can seamlessly migrate to new platforms and create new apps. Apple has prepared a new app store for Apple Vision Pro, where users can discover applications and content developed by developers for visionOS, and access thousands of compatible iPhone and iPad applications.
Eye + hand tracking, breakthrough system-level interaction. The Vision Pro headset is equipped with an external 3D camera module + Insideout spatial positioning technology to capture hand positioning, and a built-in sensor to read eye-related data, allowing users to select specific buttons, applications, or list items with only eye aims, and then pinch with their thumb and forefinger to activate the task. Currently Quest 2 only allows users to use hand tracking to control the interface, PSVR 2 only provides eye tracking technology for certain game menu selection and navigation, Quest Pro includes both eye tracking and hand tracking technology but cannot connect the two in series for interaction or selection, Vision Pro headset can achieve breakthrough system-level interaction between eye and hand tracking technology.
2.1.2. Broadening application scenarios is the focus of industry development, and AI and XR development complement each other
XR equipment upstream is mainly equipment parts, according to the cost of parts are divided into various chip units (35-40%), built-in display (25-30%), built-in lenses (10-15%), external cameras (10-15%), various structural parts (10-15%) and interactive handles (5-10%). Among them, chips, displays and optical lenses are the core links that determine the device experience.
The midstream of the industry chain is an OEM integration solution provider and terminal equipment brand. Mainstream VR/AR device brands include Meta, PICO (byte), Dapeng, HTC and iQiyi. The XR device landscape is highly concentrated, and the market share of Oculus XR devices launched by Meta has remained above 50% for a long time since Q4 2020, followed by PICO. The downstream aspects of the XR industry chain are divided into VR/AR ecological content producers and publishing platforms that provide ecological content. Upstream equipment hardware and midstream terminal equipment manufacturers innovate according to the application needs of downstream customers, and downstream customers are divided into B-side (consumer-level application market) and C-side (enterprise-level application market). B-side application scenarios are mainly divided into five areas: video, live broadcast, gaming, education, and social networking. Overall, games accounted for 40.5% of VR content in 2020, followed by video (20.1%), live streaming (8.9%), healthcare (8.6%), and education (7.7%).
AI and XR developments complement and reinforce each other. On the one hand, AI technology can collect all kinds of data on XR devices and process it efficiently, on the other hand, the development of AIGC will further improve the efficiency of content production. The development of XR devices has reversed the integration of AI technology and industry.
From a data processing perspective, Meta recently released a new AI model, Segment Anything Model (SAM), which can produce masks for any object in any image or video, even including objects and image types that have not been encountered during training. SAMs are versatile enough to cover a wide range of scenes and can be used out-of-the-box on new image domains without additional training. The SAM model opens up a new app space for AIGC apps. Through research and dataset sharing, Meta hopes to further accelerate research into image segmentation and more general image and video understanding. Meta expects composable system designs based on techniques such as prompt engineering to support a wider range of applications than systems trained specifically for a set of fixed tasks. SAM can be a powerful component of AR, VR, content creation, the scientific field, and more general AI systems.
From the perspective of rich application scenarios and ecological improvement, the integration of AI and VR/AR technology will further enrich virtual content and improve R&D efficiency. 3D content (including 3D models, 3D animation, and 3D interaction, etc.) is one of the cores of VR/AR, and the current 3D content in various fields still requires a lot of manual production, and the threshold requirements for production personnel are relatively high, so the production capacity is low, which is a major bottleneck restricting the development of related industries. AI is expected to automate 3D content production to a certain extent, replace some repetitive work, and improve production efficiency. At present, AIGC has demonstrated strong production capacity in the field of text, pictures, music and other content generation, compared with AI-generated text, images and videos, AI-generated 3D content is still in its infancy. However, with the development of AI models and the improvement of computing power, AI is expected to bring about changes in the field of 3D content generation. At present, the main form of AR/VR content is games, AI and VR/AR games combined, by constantly using human dynamic data, language, reactions and other information to train AI, VR/AR games in virtual character movements, behaviors or more smooth, and more intelligent, thereby greatly enhancing the immersion of the game. From the perspective of terminal hardware access, on April 11 this year, brands of Innovative Eyewear, a cutting-edge smart glasses manufacturer, announced the launch of a new iOS/Android application called Lucyd, which supports ChatGPT voice interface, and users can ask ChatGPT questions through the built-in microphone of glasses and hear answers through stereo speakers, becoming the first smart glasses company to provide ChatGPT. Jiahe Intelligent also said that the smart glasses that the company has shipped to customers have been equipped with its AI voice assistant model, and customers will also carry their own AI large models according to their planning, and the company's products will leave corresponding interfaces for customers to choose to access AI large models, and there is a relevant layout in product planning.
2.2. Mobile phone: It is expected to become the most shipped and interactive AI hardware entrance
2.2.1. The demand for mobile phones is slowing down, and technological innovation may become an important driving force to stimulate replacement
In Q1 2023, China's smartphone shipments fell 8% year-on-year, narrowing from the year-on-year decline in previous quarters. IDC expects China's mobile phone market shipments to decline by 0.9% year-on-year in 2023, and the decline will continue to narrow, and it is expected to usher in a rebound in 2024. Overall, mass-market demand remains weak, with all major Android phone makers posting month-over-month declines, cautious in ramping up production, and weak demand management to avoid any unhealthy inventory overhang in the market. Apple's increase in market share lies in the launch of the iPhone 14, further expanding into China's high-end mobile phone market.
Smartphones have excess performance and are in dire need of software or application updates with obvious breakthrough advantages. According to a report released by Strategy Analytics, there are five behavioral patterns in the decision of users to replace their mobile phones: brand lovers, low-cost practicals, status-hungry people, technology enthusiasts, and value maximizers. Among them, the highest proportion of value maximizers is 25%, followed by technology enthusiasts at 24%. It can be seen that under the premise of eliminating the added value of the brand, the mobile application experience with a higher level of innovation has become an important source of power to promote users to replace mobile phones.
2.2.2. AIGC-enabled smartphones are expected to become a new demand driver
In the context of the gradual narrowing of hardware innovation, AIGC-enabled smartphones are expected to become a new demand driver.
On May 19, 2023, Open AI announced on its official website that it will launch the ChatGPT app for iOS in the United States, which will be expanded to other countries in the coming weeks. The app provides users with customized suggestions, inspires creativity, or enables personalized learning. In addition, Open AI also said that it will soon launch a ChatGPT app for Android.
On May 11, 2023, Google released Android 14 at the I/O conference, introducing magic compose (used in messages and conversations, rewriting text in different styles, modifying language through AI), generative AI wallpapers (Generative AI Wallpapers, you can choose a variety of styles and keywords to generate high-definition pictures) and other functions, it is expected that AI technology will further empower smart phones in the future. In turn, it accelerates the popularization and application of smart phones.
At present, the AIGC field presents trends such as continuous enrichment of content types, continuous improvement of content quality, and increasing versatility and industrialization of technology, and applications such as writing assistants, AI painting, dialogue robots, and digital humans have emerged, supporting content demand in media, e-commerce, entertainment, film and television, and other fields. In the future, with the improvement of smartphone hardware computing power, related AIGC applications may be popularized in smartphone terminals.
Taking Transsion Holdings as an example, the company released the TECNO Camon20 in May this year, which is the world's first mobile phone to access OpenAI and realize ChatGPT's "text input and text output" AI assistant; At the same time, Infinix, a subsidiary of the company, launched X-bartender, an AI guide with multilingual dialogue capabilities based on AIGC technology.
TRANSSION Holdings has been laying out AI-related technologies since 2018, and established an AI specialized technology department in 2020, which currently has layouts in multilingual AI voice assistants, AI scene recognition, digital human technology, and vertical multimodal perception and interaction technologies. The company is accelerating the layout of intelligent voice technology in emerging markets, enabling multi-scenario applications of intelligent terminal products and mobile Internet services, and the "African native language AI speech technology pioneer" project jointly led by the company's AI technology department and Tongji University won the first prize of the 2022 China-Africa Youth Innovation and Entrepreneurship Competition. In addition, the company officially launched the self-developed tens of billions of parameters large language model Sahara, currently in some scenarios of Q&A evaluation close to 92.35% of ChatGPT, at the same time the company actively carries out vertical field multi-modal perception and interaction technology and other research, based on long-term accumulation of voice, image, video and other databases for Africa, we expect that the company is expected to build a vertical field AI model beyond the general large model in the future, bringing users a better experience.
2.3. IOT Smart Terminal: AI Empowerment Expands Application Scenarios
The global Internet of Things is still growing at a high rate. According to the GSMA's "The mobile economy 2020" report, the total number of global IoT connections reached 12 billion in 2019, and it is expected that by 2025, the total number of global IoT connections will reach 24.6 billion, with a compound annual growth rate of 13%. Mainland IoT connections account for up to 30% of the world's total, with 3.63 billion IoT connections in 2019. According to data at the World Internet of Things Conference in September 2021, the number of mainland Internet of Things has reached 4.53 billion at the end of 2020, and it is expected to exceed 8 billion in 2025. In 2019, the size of the mainland IoT market was about 1.76 trillion yuan, and in 2020, according to the data released by CCID, the mainland IoT market size reached about 2.14 trillion yuan. According to preliminary statistics, the market size in 2021 will be 2.63 trillion yuan. In the future, the broad market demand will bring rare development opportunities and broad development space for the Internet of Things. In the next 6 years, the development of China's Internet of Things will maintain rapid growth, and the market size will exceed 7 trillion yuan by 2027.
IoT is powered by three key emerging technologies: artificial intelligence (AI): programmable functions and systems that enable devices to learn, reason, and process information just like humans. 5G network: Fifth-generation mobile network with high speed, near-zero latency for real-time data processing. Big data: Large amounts of data processed from numerous internet-connected sources. As the number of IoT devices increases, the flood of data will also be natural. This is where AI steps in – extending its learning capabilities to the IoT and enhancing their connectivity.
2.3.1. Smart set-top boxes: It is expected to become a carrier device for smart home terminals from video control terminals
As a terminal for various media assets and content calls, set-top boxes have AI landing scenarios. Based on AI functions, set-top boxes are expected to become smart home terminal carrier devices from control terminals:
1) Expand functions: by integrating near-field and far-field voice, linking to the TV screen, expanding functions such as search, education, gaming, health, and AI assistant; Through AI intelligent perception, intelligent perception of temperature and humidity, smoke perception, PM value, elderly and children, etc. is realized; Centralized control of TVs, door locks, cameras, lights, curtains, etc. through intelligent control functions.
2) Realize voice control: users control through language at home and control through mobile phones remotely. And through OpenAI's custom language model, it can answer user questions with high quality, and also collect questions to provide feedback to developers.
3) AIGC intelligent search to improve user experience: Based on AIGC, users can control terminal devices with multiple rounds of dialogue, call media asset libraries, and realize, recommend, call and present content such as film and television films, live sports, education, e-commerce, and local city information, and improve user convenience, frequency and user stickiness.
2.3.1. Smart audio: The combination of voice recognition + search engine is expected to become the control center of smart homes
Smart speakers are an upgraded version of ordinary speakers, which achieve the purpose of interacting with smart homes through voice commands. "Intelligence" lies in the combination of speech recognition + search engine to respond quickly to results. In the context of the development of voice interaction technology, smart speaker products are gradually expanding to the control center of smart homes while providing content entertainment and life services.
In the blooming smart speaker market, domestic and foreign technology giants and artificial intelligence startups have sprung in. On the brand side, Alibaba released "Tmall Genius", iFLYTEK and JD.com jointly established Linglong Technology and released "Dingtone Speaker", Lenovo launched "Lenovo Intelligence", Xiaomi launched "Xiaomi AI", and startups with Google investment background also released "Ask Speaker". At present, the competition in the smart speaker market is fierce, and Internet manufacturers dominate, the top four brands are Baidu, Xiaomi, Tmall Genie, Huawei, with a market share of 35%, 31%, 27%, and 4% in 2022, respectively, accounting for 97% of the market share of smart speakers in China.
In the assembly process, Guoguang said on the investor interactive platform on June 15 that the company's current smart speakers equipped with ChatGPT are based on the computing power of the cloud to provide feedback content. In the future, AI speakers are equipped with large models or small models, and need to match different computing power to provide solutions, the company has research and development and design solutions for various types of AI speakers.
In the chip link, the mainstream smart speaker main control chips on the market are mainly from: Apple, Quanzhi, Jingchen, Tronics, Baidu, Broadcom, Junzheng, Intel, Torch, Jieli, Lanzhi, MediaTek, Meiman, Rockchip, etc. There are not only self-developed chips such as Apple, Baidu and other terminal brands, but also chip designers specializing in audio equipment SoC main control chips such as Allwinner Technology.
ChatGPT is expected to solve the pain point of smart speakers and drive smart speakers to usher in a new round of leaps. Voice interaction is the core function and selling point of smart speakers, and traditional smart speaker products are difficult to achieve "conversational" communication between people, and it is difficult to cope with complex dialogue scenarios. At present, with the advent of new language models such as ChatGPT, whether it is reasoning and analysis of more complex languages, excellent conversation coherence and fluency, or more personalized services that can be provided, there is hope to extend to smart speakers. The arrival of ChatGPT enables smart speakers to provide users with more comprehensive, intelligent and personalized services from automatic understanding, language generation, to smooth conversations and Q&A, and the development of artificial intelligence is expected to drive smart speakers to usher in a new round of leaps.
2.3.2. Smart headsets: AI services are moving in the direction of mobility and personalization
Although it is also a smart audio product, the positioning of smart headphones and smart speaker products is fundamentally different. Hidden behind smart headsets is the trend of AI services and capabilities developing in the direction of mobility and personalization, and this on-the-go use demand will eventually give birth to highly generalized AI suitable for all scenarios. Headsets can keep users online full-time without occupying vision, which is a better supplement for people to access AI services.
On May 24, 2023, iFLYTEK officially released a new generation of recording noise reduction conference headset iFLYBUDS Nano+, which not only realizes full-scene recording to text, multilingual recording and translation, but also has VIAIM AI generative conference assistant and other functions, focusing its core functions on the office and giving the headset more practicality. VIAIM AI is an AI-generated meeting productivity tool that intelligently processes, edits, and manages text-based to meet the needs of everyday meetings. At present, functions such as summary summary extraction, to-do extraction and to-do follow-up can be realized.
AI translation is also one of the important use cases of headsets. Compared with traditional handheld translation machines and translation software, wearable AI translation headsets can truly free your hands and communicate freely in a more convenient and efficient way, making cross-language communication more natural and fluent. The application of AI technology enables translation headsets to deeply understand and analyze context, providing accurate, natural, and contextual translation results.
3. B-side: AI+ improves production efficiency, and the development potential of application scenarios is gradually released
With the development of technology and applications, the empowerment of AI will profoundly change the productivity and production efficiency of many B-ends, which can help enterprises improve production efficiency, enhance customer experience and create new business value. In the future, the continuous development and application of AI technology will bring more opportunities and challenges to enterprises.
3.1. Security: AI + security intelligent transformation to promote innovative business development
As the digital economy continues to be combined with the real economy, the security industry has taken the lead in benefiting from the mature development of the AI industry. AI makes the security industry usher in a new round of development and innovation in terms of products and technology.
3.1.1.AI Help traditional security, intelligent transformation into an inevitable trend
China Security has occupied a high share of the global market in many years of construction and development, and the market development has entered a stable and mature stage. The security industry is obviously affected by policy-driven, and safe city and snow bright engineering policies have become important policy driving forces in the security industry in the past decade. From 2015 to 2018, the large-scale promotion of the Xueliang Project, the growth rate of public safety financial expenditure in the whole country was higher than 10%; With the two major policies coming to an end, the growth rate of the security industry has entered a period of steady growth. In 2022, the domestic economy was affected by the epidemic in many places, and the slowdown in project implementation, the delay in construction period, the lengthening of the delivery cycle, the shortage of funds and the final settlement problems have suppressed the demand of the industry to a certain extent. With the optimization and adjustment of domestic epidemic prevention and control policies, all localities have once again listed economic development as the primary goal, and it is expected that security demand is expected to repair along the economic cycle from 2023.
The intelligent transformation of traditional security is an inevitable trend, and AI helps the transformation of the security industry. Since 2020, the software penetration rate in the domestic security market has increased, and the trend of project turning to intelligence has become more obvious. Intelligence has become the mainstream trend of the security industry, promoting the security market to maintain positive growth. Driven by smart projects, the scale of China's intelligent video surveillance market is growing faster than the global market, with China's intelligent video surveillance market reaching $10.6 billion in 2019, accounting for about 48% of the global market, and it is expected that China's intelligent video surveillance market will reach $16.7 billion by 2024; The 5-year CAGR is expected to be 9.6%, higher than the global market of 7.1%.
During the "Thirteenth Five-Year Plan" period, the security industry mainly developed hardware equipment such as cameras. During the "14th Five-Year Plan" period, due to the changes in the planning policy outline, the proportion of software platform and business system construction increased significantly, and the AI+ security track began to enter the era of "data intelligence". In the future, the revenue of AI+ security will be more concentrated on "video data middle platform construction" and vertical business system development.
3.1.2. Security solutions are evolving to cloud+edge solutions
The landing of AI in the field of video surveillance first starts from the cloud, that is, adding AI computing functions to back-end products to realize intelligent analysis of video data. With the improvement of the clarity of the camera, the increase of the shooting distance, the increase in resolution, and the amount of data transmitted back through the network is also increasing, the complete concentration of data structure, processing and analysis in the cloud will have problems such as network transmission pressure, real-time requirements cannot be met, and accuracy is reduced.
The "cloud + edge" edge computing solution injects AI computing power into the edge, improves the response speed of AI servers, and greatly reduces network operating costs. By adding an AI chip to the webcam, the front-end camera can structure the video data in real time. For example, cameras installed at traffic intersections can extract car information such as license plates and models, as well as passenger information such as the number of passengers and whether they are wearing seat belts, and transmit them back to the data center, which is convenient for real-time analysis and optimizes the response ability of the system.
According to the different degree of intelligence, intelligent cameras can be divided into three levels: Intelligent network camera (Smart IPC): On the basis of traditional IPC, some specific recognition functions (such as behavior analysis, anomaly detection, statistical functions, etc.) are added, without additional AI coprocessors, and are expected to take the lead in large-scale promotion. Structured analysis camera: Taking Beijing Junzheng's T20+T01 solution as an example, a simple NPU shallow processor is added to the IPC SoC, which can extract some specific information from video data, such as license plates and vehicle information. Deep learning camera: There is a separate AI chip (GPU/FPGA/ASIC), which uses deep learning algorithms to process video data in a structured manner and do analysis in different dimensions to improve the recognition accuracy of targets.
Under the trend of cloud-edge integration, front-end intelligent chips usher in greater opportunities. The key to cloud-edge integration is to analyze video analytics capabilities from full cloud to cloud terminals, so the importance of front-end AI capabilities will increase significantly. To achieve front-end AI capabilities, AI chips built into front-end devices are essential.
3.2. Machine vision inspection: AI+ improves detection accuracy and replaces traditional human eye inspection
Artificial intelligence will also be more widely applied to the intelligent manufacturing industry. The evolution of software and hardware technologies such as precision optics, computer software algorithms, mechanical motion, and electrical control has laid a deep underlying technical foundation for intelligent manufacturing. The "New Generation Artificial Intelligence Development Plan" issued by the State Council mentioned that the integration of artificial intelligence and manufacturing will be comprehensively promoted to solve the problems faced by China's manufacturing industry in the process of promoting intelligent transformation. After manual manufacturing, assembly line, automation, flexible automation and integrated automation, the form and complexity of equipment have also changed accordingly, experiencing three historical development stages of mechanization, electrification and digitalization, and intelligence has become a development trend. In recent years, the evolution of software and hardware technologies such as precision optics, computer software algorithms, mechanical motion, and electrical control has laid a deep underlying technical foundation for intelligent manufacturing. The "New Generation Artificial Intelligence Development Plan" issued by the State Council mentioned that the integration of artificial intelligence and manufacturing will be comprehensively promoted to solve the problems faced by China's manufacturing industry in the process of promoting intelligent transformation.
Machine vision as part of artificial intelligence has been widely used in intelligent equipment, through the application of deep learning, augmented learning, transfer learning and other technologies, intelligent manufacturing will improve the efficiency of knowledge generation, acquisition, application and inheritance in the field of manufacturing. Machine vision is to use machines instead of human eyes to detect and judge, mainly through the image sensor to convert the acquired target into image data, transmitted to the special image processing system, the image processing system performs various calculations on these image data to extract the characteristics of the target, and then according to the results of the discrimination to control the on-site equipment action.
Machine vision inspection is one of the best footholds for AI technology in intelligent manufacturing. For example:
Technologies such as AI image enhancement technology can preprocess images through artificial intelligence machine learning algorithms, improve the contrast of photos taken or scanned by industrial cameras, reduce image noise, enhance image edges, etc., and improve the accuracy of subsequent further inspection.
The AI defect detection algorithm can complete a large number of model pre-training for various defects such as crush injury, bump, and foreign body common in consumer electronics products. When calling the vision inspection platform in a specific project, it can use 50-100 image data to build detection parameters that meet the needs of the project through small-sample training detection technology, which has the characteristics of flexibility, can be quickly deployed to project applications, and can be adapted to a variety of defects to solve complex defect detection.
Although the intelligent equipment industry in mainland China has developed rapidly in recent years, the difficulty of gathering funds and professional talents has restricted the large-scale development of enterprises in the industry to a certain extent, and small and medium-sized enterprises are still the main enterprises in the industry. According to the prospectus of Rongqi Technology, in 2020, the company's market share in the mainland machine vision industry will be about 2%, of which the market share in the field of consumer electronics will be about 4%, and the market share will be large and the growth space will be broad.
4. Sorting out the target
The AI industry chain is long, opportunities arise frequently, the development of AI applications will be the driving force to promote the common prosperity of the upstream and downstream of the industrial chain, and the relevant listed companies will sort out:
Terminal brands: Transsion Holdings, Ramblers, Skyworth Digital, Guoguang.
B-end applications: Hikvision, Dahua Co., Ltd., TZTEK Technology, Lingyunguang.
Server-side: PCB leader Hudian Co., Ltd., global foundry leading industrial Fulian.
Chip end: SMIC, the only advanced process fab in mainland China;
Edge SOC: Rockchip, Allwinner Technology, Zhongke Lanxun, Espressif Technology, Hengxuan Technology, Jingchen Shares;
Storage: GigaDevice Innovation, Montage Technology, Juchen Co., Ltd., Beijing Junzheng.
Parts and assembly: Lingyi Intelligent Manufacturing, Zhaowei Electromechanical, Sanlipu, Goertek, Changying Precision.
Transmission end: Zhongji Xuchuang, Xinyisheng, Huagong Technology.
(Report source: Soochow Securities, for reference only, if copyright is involved, please contact to delete.) ）