Research On Agricultural Pest Survey Based On Machine Vision

As a large agricultural country,China often suffers huge losses of crops due to insect pests,so preventing pests is a prerequisite for ensuring high crop yields.Traditional pest monitoring systems often rely on the human eye to read notes,which are low efficiency and high cost.With the development of computer technology,image-based pest identification and counting methods are gradually becoming a research hotspot in agricultural pest monitoring.In this paper,based on pest image samples,a weighted decision algorithm based on color space moments and geometric features and an improved YOLOv3 algorithm are used to detect and implement pest detection and reporting based on machine vision technology.The main research contents of this article are as follows:(1)Firstly,a variety of image detection algorithms are analyzed and compared.A weighted decision algorithm based on color space moments and geometric features is used to detect pest images with a good background and a single pest type.For pest images,the background is complex The species of insect pests were detected using the improved YOLOv3 algorithm.(2)Optimize the existing pest image acquisition terminal.In view of the blurring of the pest image during the upload of the pest image,the image evaluation method is used to score the pest image before uploading.When the score is lower than the given threshold,the upload is stopped and the surveyor is notified to calibrate the collection end.Because the acquisition end is in a wild environment and there is a problem that the network signal is too weak,the DCT transform method is used to compress the pest image to relieve the network pressure when uploading.After testing,the memory space occupied by the pest images is significantly reduced.(3)Aiming at the characteristics of stable pest illumination and single sex attractant in some pest images,color space conversion is used to segment the pest images into backgrounds,and then color space moments and image geometric features are used to extract features of the pests.Identification of pests.After testing a single pest image,the recognition accuracy is 93%.(4)As the volatility time of the neutral attractant of the attractant increases,the types of pests on the sticky board will increase;at the same time,the device is in the wild for a long time and is affected by various factors,which makes the background of the pest image complex.The YOLOv3 algorithm is used to detect sticky pest images.First,the K-means algorithm based on particle swarm clustering is used to cluster the pest images to obtain corresponding anchor boxes;second,the original YOLOv3 network layer is improved,and the densely connected network is introduced to deepen the network layer to obtain more 8 image features;Finally,an improved non-maximum suppression algorithm is applied to the detection frame to penalize the high overlapping frame and reduce the missed detection rate.The final detection results show that the improved algorithm is significantly better than the original algorithm,and the average accuracy rate of pest recognition reaches 98%.In this paper,a corresponding detection algorithm is proposed for the image of sticky board pests,and the effectiveness of the proposed algorithm can be applied to more complex detection environments,and the recognition accuracy can meet the requirements of the pest detection and reporting system.