Research On Detection Method Of Diseased Leaves Based On Deep Learning

The growth of crops is very important to the development of agricultural economy.Crops are easily affected by various diseases in the growth process,which will cause serious problems such as crop yield reduction.Crop diseases usually start from the leaves and then spread to the whole plant.Therefore,efficient and accurate detection and identification of sick leaves are crucial for accurate control of crop diseases.However,at present,most pathological leaf detection models have high parameter memory requirements and slow reasoning speed,which is difficult to be implemented in agricultural production.In order to solve the above problems,based on the existing lightweight network model,an optimization scheme was proposed in this article,which was applied to the detection and identification of crop sick leaves.The main work of this article is as follows:(1)Based on lightweight convolutional neural network Mobile Net V3-Small,the ECM-Mobile Net V3-Small network model is constructed by introducing ECA attention mechanism,model compression and activation function optimization,and multi-scale feature fusion structure design.To realize the efficient identification of leaf disease.The experimental results show that the model can not only guarantee the accuracy,but also greatly reduce the number of network parameters and improve the calculation speed of the model.The average recognition accuracy reaches 99.54%,the model training time is about 3.73 h,the recognition time of a single image is 27 ms,and the number of parameters is about 0.29 Mi B,which is only 3/10 of the original model.Compared with Mobile Net V3-Small and other classical network models,it shows better performance.(2)In view of the multi-target,multi-disease and complex background of crop pathological leaves in natural environment,a model structure based on Yolo V4-tiny was proposed,combined with ECM-Mobile Net V3-Small convolutional neural network.The detection model of ECM-Mobile Net V3-Yolo V4-tiny sick leaves was constructed.At the same time,a deep separable convolution structure is introduced to reduce the number of network parameters and the amount of computation,and a Res-SPP structure is designed to increase the network receptive field by combining SPP and residual network ideas.The m AP of ECM-Mobile Net V3-Yolo V4-tiny model is about 93.48%,the number of parameters is 0.39 Mi B,the floating point computation is about 0.71 G,the model training time is about 1.15 h,and the average detection time of a single image is about 39 ms.Compared with Yolo V4-tiny and other classical target detection algorithms,the model has a light structure,fast reasoning speed,and can guarantee high accuracy,with good performance advantages.In summary,the detection and recognition method of crop morbid leaves proposed in this article can well realize the efficient and accurate detection and recognition of crop morbid leaves.Besides,the model structure is light,which can provide technical reference for the development of mobile detection system of crop morbid leaves and has certain application value.