Authors

Abstract

Early diagnosis and accurate identification of fine-grained tomato diseases can control the spread of diseases and insect infections, thus ensuring the healthy development of the tomato industry. In this paper, four lightweight models of Tiny-AlexNet and Mid-AlexNet based on AlexNet and Tiny-VGG16 and Mid-VGG16 based on VGG16 were proposed for 5 kinds of early and late leaf diseases such as tomato powdery mildew. The computation speed of the model is accelerated by reducing the number of neurons in the fully connected layer. In order to avoid degradation in network training, data extension technology is introduced to prevent model overfitting. Among them, the Mid-VGG16 model is significantly better than accurate in early disease recognition, thus verifying the effectiveness of the lightweight model. The proposed model not only improves the accuracy, but also reduces the test time. The results were tested across 20 655 data sets on early and advanced disease. Compared with the traditional model, the average prediction accuracy of the proposed model is improved by about 0.15%, and the detection time is significantly reduced by about 50%. The improved model has strong robustness and high stability. The model can be used to accurately identify early diseases and facilitate real-time detection and prevention of tomato diseases.

Abstract in Chinese

细粒度番茄病的早期诊断和准确识别可以控制病虫害感染的传播，从而确保番茄产业的健康发展。本文针对番茄白粉病等5种早期和晚期叶子病，提出了基于Alexnet的Tiny-Alexnet、Mid-Alexnet和基于VGG16的Tiny-VGG16、Mid-VGG16 4种轻量级模型。通过减少全连接层中神经元的数量，加快了模型的计算速度。为了避免网络训练中的退化，引入了数据扩展技术，防止了模型的过拟合。其中Mid-VGG16模型在早期疾病识别方面的准确率明显优于晚期，从而验证了轻量级模型的有效性。提出的模型在确保准确精度提高的同时，测试时间的消耗也大大减少。实验结果通过20 655 张关于早期和晚期疾病的数据集测试，与传统模型相比，提出的模型的平均预测准确率提高了约0.15%，检测时间得到了约50%的显著降低。经过改进的模型具有较强的鲁棒性和高稳定性。模型可用来准确识别早期疾病，进而便于番茄疾病的实时检测与防治。