Research On The Method Of Apple Leaf Disease Detection Based On SSD Network

The apple industry is an important part of the economic development of agricultural products in Shaanxi Province.However,its growth process is affected by various diseases,detecting apple leaf diseases accurately can prevent diseases from spreading and promote the healthy growth of the industry.The existing algorithms cannot achieve accurate detection of apple leaf diseases due to problems such as varying disease spot sizes and weak location information.Also,due to the difficulty of collecting them,it is difficult for the few shots of new classes to meet the training requirements of traditional object detection algorithms.To solve the above problems,this paper proposes an apple leaf disease detection method based on SSD(Single Shot Multi Box Detector)network for nine common apple leaf diseases in Shaanxi Province,and the specific research contributions are as follows.(1)A multi-scale feature fusion method for apple leaf disease detection is proposed.To solve the problem that traditional algorithms cannot effectively adapt to multi-scale detection tasks due to the different sizes of apple leaf diseases,this paper optimizes the feature extraction capability of the model for multi-scale disease spots by feature fusion techniques and proposes a V-space-based multi-scale feature fusion SSD(VMF-SSD).Firstly,multi-scale feature extraction is established to fuse features at different levels to further improve the detection performance of apple leaf diseases,especially small spots.Then the V-space localization branch is proposed,which plays an important role in enhancing the texture feature information for disease spot localization.Meanwhile,an attention mechanism is used to automatically learn the importance of feature channels at different scales to distinguish among different sizes of disease spots.The experimental results show that the VMF-SSD method achieves 83.42% mAP on the test set and obtains a detection speed of 27.53 FPS.Compared to the existing object detection algorithms,the algorithm can balance accuracy and speed simultaneously.And the proposed method for apple leaf disease detection can achieve better performance in the apple leaf disease detection task and meet the requirements of practical applications in agricultural production.(2)An apple leaf disease detection method based on VMF-SSD for few shots is proposed.To solve the problem that the traditional object detection algorithms cannot guarantee effective convergence due to the few apple leaf disease samples,this paper optimizes the feature learning ability of the model for few shots by reconstructing the loss function,and proposes a SSD method based on contrast embedding loss and IoU-CE loss(CI-SSD).Firstly,it introduces the weight parameter of VMF-SSD to enhance the representation ability of the feature extraction network.Then the prior box decoding module is constructed to generate more robust potential feature representations,and contrastive embedding loss function is modeled to constrain the similarity between different classes of prior box,thus reducing the accuracy loss caused by misclassification.Finally,by analyzing the distribution pattern of samples and IoU,the classification loss function based on IoU and Softmax is constructed to enhance the loss weight of positive samples in the classifier and further improve the detection accuracy.The experimental results show that the accuracy of CI-SSD method is 70.53% mAP and 40.92% mAP for the base class and new class,respectively,in the case of 10 shots.Compared with existing algorithms,the proposed few shot disease detection method has strong generalization ability on new classes of diseases,which provides a new way to explore new methods for other crop diseases detection.