Research On The Detection Technology Of Silkworm Diseases And Pests Based On Deep Learning

As an important economic activity in many south areas,sericulture has become the main way to achieve rural revitalization in Guangxi.In the breeding process of small sericulturists,it is easy to be affected by various factors such as the environment to appear sick silkworms,and silkworm disease has a certain degree of contagiousness,if the sericulturist can not accurately determine the type of sick silkworms and make the corresponding management,it will affect the economic efficiency.The main difficulties in the detection of silkworm disease are the following: such as the similarity between different silkworm diseases,the image is easily affected by the environment when taking pictures to identify silkworm diseases,and multiple diseased silkworms are difficult to detect at the same time.To address the problems in silkworm disease detection,this paper proposes research on deep learning-based silkworm pest detection technology from two aspects: improved image detection segmentation technology and deep learning model,which is used to solve the problem of auxiliary detection of diseased silkworms when silkworm farmers breed,reduce the misjudgment of diseased silkworms and improve the intelligent development of silkworm breeding industry.The main research work accomplished in this paper is as follows:(1)To address the problem of background removal in the pre-processing stage of diseased silkworm images,we propose to use the component features of the images combined with the peak density of the data to recalculate the clustering centers of the K-means algorithm and automatically determine the image segmentation thresholds.By comparing the commonly used threshold segmentation algorithms with the proposed method,it can be seen that the proposed method can effectively remove the background area of the diseased silkworm image with an accuracy of 85%,which has obvious advantages compared with other image segmentation methods.(2)The model structure is improved based on the YOLOv5 algorithm.For the case of sample imbalance with similar characteristics of some silkworm diseases,the SPPCSPC module is added behind the C3 structure in the backbone part of YOLOv5,and the Shuffle Attention mechanism is added in the detection head part.Then the original dataset mixed with diseased silkworm images preprocessed by image background removal is used to train the improved YOLOv5 model,to improve the model’s learning of diseased area features.The improved model structure has higher accuracy and computational speed compared with the original structure through comparison experiments.In this paper,we improve the image detection segmentation technique and YOLOv5 model structure in two directions,and the detection model for common silkworm diseases meets the expected objectives in terms of accuracy and speed,which can assist in silkworm disease detection.