Application Of Object Detection Algorithms Based On Deep Learnining In Detection Of Stored-grain Pests

As an important material for the development of human society,grain plays an important role in the social stability of the country and even in the peace and development of the world.China is a big country of grain production,but the damage of stored grain pests in the process of grain storage is very serious.There are many methods to detect grain pests.In this paper,object detection technology based on deep learning is used to detect and identify stored grain pests.The detection of stored grain pests under white board background and actual background is studied respectively.Different algorithms are used to carry out a lot of experiments,and the detection and recognition effect of stored grain pests is constantly improved by improving the model.The main research work of this paper is as follows:1.Comparing the traditional image recognition method and the object detection method based on deep learning,this paper deeply studies the structure and design of deep convolution neural network,and the framework,principle and different architecture of object detection algorithm based on deep convolution neural network,which lays a foundation for the detection and identification based on deep learning of grain storage pests.2.A large number of experiments with open source model were carried out on the obj ect detecti on datasets consisting of eight kinds of stored-grain pests under the background of white board.The advantages and disadvantages of two-stage detection model represented by Faster R-CNN and R-FCN and single-stage detection model represented by YOLO are analyzed,and the methods to improve the detection effect of stored grain pests are explored.On this basis,the two-stage detection model R-FCN is improved.Furthermore,data augmentation,multi-scale training and soft-NMS are used to improve the object detection effect of stored grain pests.3.This paper visualizes the convolution features of VGG16 in different network depths,and the characteristics of feature extraction with different depth of deep convolution neural network are explored.Besides,several different ways of using convolution features and the more efficient ways of using features for object detection tasks are studied.4.In the stored-grain pests object detection dataset under actual background,the background of image is complex and the occlusion problem is serious.Through the analysis of the utilization mode of convolution features,FPN,which is the feature fusion model based on Top-down structure,is introduced to detect grain insects under the actual background.Besides,the detection effect of grain storage pests under the actual background is improved by means of data augmentation,multi-scale training and soft-NMS.