Detection Of Pests In Granary Based On Deep Learning

China is the largest producer and consumer of grain,but there are still some problems in the storage of grain in China,such as breeding pests due to the impact of storage temperature and humidity,and damaging grain production.It is the first step to detect and identify the pests in the granary.In the past,the speed and efficiency of artificial recognition were very limited,so it is urgent to detect pests in granary by means of computer vision and image processing.In this paper,based on the deep learning algorithm,the image of grain warehouse pests is detected.The goal is to detect the common domestic grain warehouse pests in the background of grain warehouse.The main research contents and achievements of this paper are as follows:(1)In this paper,six kinds of common granary pests are selected,and the image data set of sgi-6 granary pests is constructed by the way of network search and download and laboratory shooting.In order to solve the problem that the information obtained from the small target is small,this experiment adds the image of grain bin pests obtained from the network as the large target scale image in the training set,and adds the image taken under the laboratory microscope as the intermediate data set.The granary pest data set covers 2558 images from different shooting angles.In order to simulate the local occlusion and angle difference of pests in the real scene of the granary,and to enrich the training set,prevent the model from over fitting,and promote the model better,this paper enhances the training set by clipping and flipping to expand to 10 times of the original.The data set sgi-6 is divided into three levels(data set 1:composed of network acquired image and SLR photographed image,data set 2:composed of network acquired image and SLR photographed image and microscope photographed image,data set 3:image enhanced data set)for model training.The test results show that the performance of the model is significantly improved by adding the intermediate data set.M The highest AP was 95.24%.Another difficulty of pest detection in granary is the complex background of granary.Most of the pests in granary are mixed with grain,which increases the difficulty of pest detection in granary.In this paper,the images of rice,millet and white paper are used to test respectively.The results show that the model can be applied to the detection of pests in Granary under complex background.(2)In order to solve the problem of poor adjacent detection results,a Faster R-CNN model based on clustering features is proposed to detect the target of the image of the insect pests in the granary,which reduces the redundancy of candidate frames and the error of framing.Clustering algorithm is used to calculate the length-width ratio of the target in the image,and the best candidate frame length-width ratio is 0.67,0.99,1.23.Candidate frames of 512~2 size are removed.Two sizes of 128~2 and 256~2 are retained,and six different candidate frames are generated.Using Faster R-CNN training model based on clustering features,the final model's mAP reached 96.63%.At the same time,the number of pests in the image can be counted,which plays an early warning role in the storage of the granary.In this paper,under the background of the pest problems in the storage process of the granary,the pest image of the granary is detected by the method of deep learning.The experimental results show that the faster r-cnn model based on clustering features is feasible to train multi-scale data sets,which solves the problem of detecting pests in granary,it can also count the number of pests in the granary,and can play an early warning role in the problem of pests stored in granary.