Volume 2,Issue 7
基于轻量化深度卷积神经网络的番茄叶片病害分类识别研究
由于传统深度卷积神经网络（CNN）模型参数冗余高、计算资源消耗大，难以满足农业无人机及边缘计算设备的实时推理需求，本研究系统分析和验证MobileNetV3、ShuffleNetV2、GhostNet 与EfficientViT 等典型轻量化模型在番茄叶片病害细粒度分类任务中的性能表现。针对包含10类病害特征的番茄叶片统一数据集，从分类精度、计算复杂度（FLOPs）、推理延迟及模型可解释性等多维度进行量化评估。实验结果表明，EfficientViT 模型在综合性能上表现最优，F1-score 值达到0.95，展现出对复杂纹理病斑的卓越特征提取能力。而ShuffleNetV2则凭借最小的FLOPs（0.15G）与最高的推理吞吐量（5625images/s），成为实时检测场景的首选方案。GhostNet 虽然参数量极致压缩，但在处理复杂病害特征时存在一定的精度瓶颈。Grad-CAM 可视化分析进一步证实，EfficientViT 能够更精准地聚焦于病灶区域，具有更强的抗背景干扰能力。本研究为智慧农业装备中的边缘端模型选型提供了科学的理论依据与数据支撑。
[1] BILLAH M M, SULTANA A, AFTAB S R, et al. Leaf disease detection using convolutional neural networks: A proposed model using tomato plant leaves[J]. Neural Computing and Applications, 2024: 1-11.
[2] SHARMA R, NAAZ S, VAIDYA P. Harnessing deep learning with AlexNet for tomato leaf disease detection in the Indian Himalayan terrain[J]. Journal of Electrical and Computer Engineering, 2025, 2025(1): 2807347.
[3] GUHA S P, AMIN A B, ARPA S, et al. A real-time application-based convolutional neural network approach for tomato leaf disease classification[J]. Array, 2023, 19: 100313.
[4] 王晨. 基于轻量级卷积神经网络的番茄病虫害分类研究[D]. 长春: 吉林农业大学, 2023.
[5] HE Z, TONG M. LT-YOLO: A lightweight network for detecting tomato leaf diseases[J]. Computers, Materials & Continua, 2025, 82(3): 4301-4317.
[6] ABDULLAH A, AMRAN A G, TAHMID A M S, et al. A deep-learning-based model for the detection of diseased tomato leaves[J]. Agronomy, 2024, 14(7): 1593.
[7] NI S, JIA Y, ZHU M, et al. An improved ShuffleNetV2 method based on ensemble self-distillation for tomato leaf diseases recognition[J]. Frontiers in Plant Science, 2025, 15: 1521008.
[8] 杨进进, 张文慧, 王哲. 基于改进的ResNet18 模型识别番茄叶片多种病害[J]. 现代计算机, 2024, 30(9): 30-34.
[9] 侯文慧, 龚昌智, 曹文昊, 等. 基于超分辨率增强与改进YOLOv8的番茄叶片病害检测[J]. 农业工程学报, 2025, 41(16): 211-220.
[10] YAN C, LI H. CAPNet: Tomato leaf disease detection network based on adaptive feature fusion and convolutional enhancement[J]. Multimedia Systems, 2025, 31(2): 178.
[11] JIAPING J, SHUFEI L, CHEN Q, et al. A tomato disease identification method based on leaf image automatic labeling algorithm and improved YOLOv5 model[J]. Journal of the Science of Food and Agriculture, 2023, 103(14): 7070-7082.
[12] SUN Y, NING L, ZHAO B, et al. Tomato leaf disease classification by combining EfficientNetV2 and a Swin Transformer[J]. Applied Sciences, 2024, 14(17): 7472.
[13] 宋国柱, 黄文静, 崔帅帅, 等. 基于改进YOLOv8n 的轻量化番茄叶片小目标病害识别方法[J]. 农业工程学报, 2025, 41(10): 232-242.
[14] SLADOJEVIC S, ARSENOVIC M, ANDERLA A, et al. Deep neural networks based recognition of plant diseases by leaf image classification[J]. Computational Intelligence and Neuroscience, 2016(6): 1-11.
[15] CHEN M, WANG C, LIU C, et al. Tomato leaf disease detection method based on improved YOLOv8n[J]. Scientific Reports, 2025, 15(1): 25837.
[16] ZAHID U, NAJAH A, MONA J, et al. EffiMob-Net: A deep learning-based hybrid model for detection and identification of tomato diseases using leaf images[J]. Agriculture, 2023, 13(3): 737.
[17] ALAMPALLY S, CHIRANJEEVI M. A smart solution for tomato leaf disease classification by modified recurrent neural network with severity computation[J]. Cybernetics and Systems, 2024, 55(2): 409-449.
[18] PREETI B, SHARMA R J, KETAN K. TomConv: An improved CNN model for diagnosis of diseases in tomato plant leaves[J]. Procedia Computer Science, 2023, 218: 1825-1833.
[19] KUMAR V, KAUR M, MITTAL V, et al. TLD-FDL: Tomato leaf disease classification using end-to-end fusion of deep learning models[J]. SN Computer Science, 2025, 6(8): 933.
[20] 许悦, 陈琳. 基于改进YOLO v8 的番茄叶片病害检测算法[J]. 江苏农业科学, 2024, 52(17): 192-200.