ꔷ Citation format ꔷ
ZHOU Qiao-li, MA Li, CAO Li-ying, YU He-long. Tomato Leaf Disease Identification Based on Improved Lightweight Convolutional Neural Network MobileNetV3[J]. Smart Agriculture, 2022, 4(1): 47-56.
ZHOU Qiaoli, MA Li, CAO Liying, YU Helong. Identification of tomato leaf diseases based on improved lightweight convolutional neural networks MobileNetV3[J]. Smart Agriculture, 2022, 4(1): 47-56.
Click DIRECT TO CNKI to read
Click directly to the official website to read (full text, free)
Tomato leaf disease identification based on improved lightweight convolutional neural network MobileNetV3
ZHOU Qiao-li, MA Li*, CAO Li-ying, YU He-long*
(College of Information Technology, Jilin Agricultural University, Changchun 130118, Jilin Province, China)
Abstract:Timely detection of tomato disease can effectively improve the quality and yield of tomatoes. In order to achieve real-time damage-free detection of tomato diseases, this study proposes a classification and identification method for tomato leaf diseases based on improved MobileNetV3. Firstly, mobileNetV3, a lightweight convolutional neural network, was selected, pre-trained on the Image Net dataset, and the shared parameters obtained by the pre-training were migrated to the model for tomato leaf disease recognition and fine-tuned. The same training method was used to transfer and compare the three deep convolutional network models of VGG16, ResNet50 and Inception-V3, and the results showed that MobileNetV3 had the best overall learning effect, and the average test identification accuracy of 10 tomato diseases under mixed enhancement and focal loss loss function reached 94.68%. On the basis of transfer learning, the MobileNetV3 model was continuously improved, the cavity convolution and perceptron structure were introduced in the convolutional layer, and the GLU (Gated Liner Unit) gate mechanism activation function was used to train the best tomato disease identification model, the average test recognition accuracy was 98.25%, the data scale of the model was 43.57 MB, and the detection time of a single tomato disease image was only 0.27 s. After 10-Fold Cross-Validation, the robustness of the model is good. This study can provide theoretical basis and technical support for the real-time detection of tomato leaf diseases.
Keywords: tomato disease recognition; convolutional neural network; transfer learning; MobileNetV3; activation function; recognition classification
Image of the article
Fig. 1 Comparison of original and augmented images of seven-star leaf spot disease in tomato leaves
Fig. 1 Comparison of original image and augmented images of tomato septoria leaf spot
Figure 2 Tomato leaf disease image Mixup enhancement
Fig. 2 Mixup enhancement of tomato leaf disease image
Figure 3 Flowchart of tomato leaf disease identification based on transfer learning
Fig. 3 Flow chart of tomato leaf disease identification based on transfer learning
Figure 4 Multilayer perceptron structure diagram
Fig. 4 Multilayer perceptron structure
Figure 5 Schematic diagram of cavitational expansion
Fig. 5 Schematic diagram of dilated convolution expansion
Figure 6 Structure diagram of the improved MobileNetV3 network model
Fig. 6 Structure diagram of improved MobileNetV3
network model
Figure 7 Comparison of four types of transfer learning: MobileNetV3, VGG16, ResNet50, and Inception-V3
Fig. 7 Comparison of the four transfer learning of MobileNetV3， VGG16， ResNet50， Inception-V3
Figure 8 Four algorithms identify tomato disease results
Fig.8 Tomato leaf diseases recognition results using the four algorithms
Figure 9 Migration MobileNetV3 model improvement before and after test curves
Fig. 9 Test curves before and after model improvement of migrate MobileNetV3
Figure 10 Improved MobileNetV3 model confusion matrix diagram
Fig. 10 Improved confusion matrix of MobileNetV3
Source: Smart Agriculture (Chinese and English), No. 1, 2022
Please contact the editorial board for authorization
Supporting units for this issue
Jinglan Yunzhi Internet of Things Technology Co., Ltd
Zhejiang Zhenshan Technology Co., Ltd
Weichai Revo Heavy Industry Co., Ltd
small
shop
Welcome to shop
WeChat communication service group
In order to facilitate the academic exchange of readers, authors and review experts in the field of agricultural science, promote the development of smart agriculture, and better serve the majority of readers, authors and reviewers, the editorial department has established a WeChat exchange service group, and the discussion of issues in the professional field and the issues related to submission can be consulted in the group.
How to join the group: add Xiaobian WeChat 331760296, Remarks: name, unit, research direction, Xiaobian pull you into the group, institutional marketing advertisers do not disturb.
Publication of Information
Introduction of scientific research team and promotion of recruitment information, academic conferences and related activities