Research On Pest Identification Based On Improved SSD Algorithm

As an agricultural country,China’s agricultural development is closely related to people’s lives and national security.In recent years,frequent pest disasters have brought enormous pressure to China’s agricultural production.Real-time and accurate monitoring of the dynamic changes of pests is of great practical significance for achieving effective pest control and promoting the high-quality development of agricultural products in China.In order to solve the problem of automatic identification of crop pests and meet the needs of high-precision and real-time pest monitoring,research on automatic detection and identification technology of crop pests based on deep learning technology has been carried out.Specifically,it includes the following aspects:(1)Establish a pest identification dataset.Aiming at the problem that there is currently no effective pest species identification dataset at home and abroad,referring to the existing open pest dataset,a multi pest identification database suitable for pest species identification is established and applied to pest species identification.By comparing two sets of photos under field and field test conditions,three main pest types were obtained: aphids,wheat spiders,and rice planthoppers.The establishment of this method provides a reference for the establishment of data sets for other exploration missions.(2)Using SSD algorithm to identify pests.After comparing and analyzing several existing detection methods,the SSD algorithm with higher detection accuracy and faster detection speed is selected and optimized.In order to solve the problems of small targets,sample adhesion,and overlap,this paper proposes a new SSD method from the perspective of network structure and a priori frame.In network construction,for small target detection,three large-scale feature maps are selected for prediction,and a fusion model of multiple features is constructed to supplement the shortcomings of large-scale feature maps in ability extraction.In the design of prior frames,the K-means method is used to perform aspect ratio matching and scale clustering on actual boundary frames,providing an appropriate prior frame size for existing detection targets,and avoiding missed or erroneous detection caused by adhesion or overlap.(3)Conduct experimental research and theoretical analysis.The correctness and effectiveness of the algorithm are verified through the analysis of practical application results.Through experiments,it was found that the optimized SSD algorithm improved the detection accuracy of m AP from 86.62% to 91.76%,while the detection time was not significantly different from the SSD method.This algorithm has a good detection effect on small targets,target adhesion,and overlapping situations in pest identification,and can meet the requirements of high-precision and real-time pest monitoring.The experimental results show that the optimization of network construction improves the recognition ability of small objects,obstructions,and small targets in complex environments;The computational performance of the algorithm is improved for dense,overlapping,and adhesive targets;This algorithm can effectively eliminate erroneous detection objects,reduce error detection practices,and also obtain more accurate boundary boxes.(4)Complete the design of the pest identification system,and after completing the requirements analysis,achieve the design and implementation of the system.The results show that it has a relatively beautiful page effect and recognition efficiency.The analysis of SSD algorithm proves the effectiveness of this method.With sufficient data volume,a more complete pest dataset can be established,which can detect more pests,provide reliable data support for agricultural pest monitoring and early warning,help scientific prevention and control of crop pests,and promote high-quality development of agriculture.