Research On Video Detection-based Assessment Algorithms Of Stored Grain Pests

In order to improve the effectiveness and economy of pest control measures,how to evaluate the pest control effect in grain depots is a problem that needs to be solved urgently.At present,the grain depot in our country fumigates insecticides by phosphine and other fumigants,so as to achieve the purpose of pest control.However,the evaluation of fumigation effect is to monitor the death of pests in the fumigation process by means of multiple artificial sampling.This method is difficult to guarantee the safety of grain depot staff and the real-time monitoring,so there is an urgent need for a real-time online automatic monitoring of the survival of stored grain pests in the fumigation process.For the moment,as a result of some grain depots in China have installed high-performance image acquisition devices for stored grain pests,these acquisition devices can take video data for us to collect the death of pests in the process of fumigation and insecticidal insecticide in grain depots.Therefore,we can design an insect cage with built-in image acquisition device,which can let the fumigation vapor pass through.First,the live insects are placed in the cage,then the cage is placed in the fumigated granary.Then,the video of the survival process of the insects in the cage is captured by the image acquisition device placed in the cage.After uploading the video to the background system,the computer vision method is used to automatically distinguish the death and live situation of the insects in the video.Therefore,this paper proposes a video detection-based assessment algorithm for stored-grain pest mortality,which can be used to monitor the changes of the specific number of pest deaths in the process of insecticidal control,so as to achieve the purpose of real-time monitoring of the changes in the number of stored-grain pest deaths during fumigation.In order to study the evaluation algorithm of stored grain pests'death and life based on video detection,this paper established a video data set to train the detection model by shooting the video of pests'death process in the laboratory environment,through which to complete the algorithm preparation for the actual monitoring of stored grain pests'death and life.The core of this algorithm is a dual-stream network based on deep convolution neural network,which integrates image target detection algorithm and two-frame difference method to identify pests in video data.The test results show that the proposed algorithm can effectively detect the survival of stored grain pests,and the average detection accuracy can reach 89.9%.The main research work in this paper is as follows:1.In this paper,a video data set of the death process of stored grain pests is established by shooting video data with Superyes camera and millet mobile phone camera.After manual marking,data expansion and data preprocessing,the data set is applied to the video detection algorithm of dual-stream neural network.2.In this paper,a video detection algorithm based on two-stream neural network is implemented to detect the survival of pests.Among them,the research of dual-stream video detection algorithm includes:the flow design of dual-stream video detection,the design of space-flow network and time-flow network in dual-stream video detection network,the best fusion method and fusion location of spatial and motion characteristics of pests.3.In this paper,based on the results of pest video target detection,a simple online and real-time tracking algorithm(SORT)is applied to the multi-target tracking task of pests,which achieves the multi-target tracking of stored grain pests and shows the pests'survival more clearly,thus verifying the detection results of dual-stream video target detection.