Research And Implementation Of Common Crop Pest Identification Method Based On Deep Convolutional Neural Network

Crops are affected by external factors during growth,which may lead to various diseases.It is very important to be able to diagnose crop diseases early so that farmers can take timely countermeasures to avoid further losses.Based on this practical need,this paper proposes a deep convolutional neural network based method for identifying common crop diseases and pests,which can identify common pests and diseases such as apple blasts and strawberry brown spots,etc.A software platform for crop disease and pest identification is designed and implemented.This thesis first introduces the research background and significance of the topic and discusses the current status of pest and disease research at home and abroad,including traditional computer vision-based pests and diseases,learning-based severe pests and conditions,and the current research status of identification and detection systems.Secondly,deep learning algorithms are studied,including a combination of traditional neural network algorithms,convolutional neural network algorithms,deep learning optimization algorithms,and other algorithms for the study and comparison experiments of neural network models for the subsequent research to be carried out and conducted.Finally,this thesis focuses on constructing and improving the pest detection and identification model.Through comparative analysis,YOLOv5 s is selected as the detection model.First,K-Means++ is introduced to obtain the target frame so that the detection results can better fit the blade.Second,to reduce the network model’s burden,the C3 module is optimized,and the PANet in the Neck module is changed to Bi FPN;Third,the loss function is improved to accelerate the convergence speed.In the end,the detection model ensures accuracy and improves detection efficiency,with m AP reaching 92.36%.The Inception-ResNet-v2 model was selected in the network model for pest species identification to ensure better identification effects for various leaf diseases.First,the SENet attention mechanism module was introduced to improve the network model and the accuracy of pest species identification,aiming at the problem of shadow and other influencing factors on the photographed leaves.Second,deep separable convolution is used to replace standard convolution to reduce the number of parameters in the network model and improve computational efficiency.Third,the improved loss function makes the classification results more accurate.In the identification data set constructed in this paper,compared with other identification networks,the accuracy rate of the improved Inception-ResNet-v2 model is 97.31%.The platform construction uses the popular Django framework,making the research results more practical.The experimental results show that this pest identification software meets the needs of actual pest identification.This study can further promote the project of crop disease detection and identification in image processing.