Research On Multi-Objective Greenhouse Pest Identification Based On Deep Learning

With the research and development of information technology and artificial intelligence application,graphic image recognition technology and target detection algorithm are widely used in industry,commerce,agriculture,medical and other areas of people’s livelihood.In the field of agricultural development,using image recognition algorithm and target detection technology to judge and detect agricultural diseases and pests has become a cutting-edge technology and research hotspot in the field of agricultural science research.Among many research directions,the combination of artificial intelligence technology and object detection algorithm has become the most popular hot issue at present.In this paper,the object recognition algorithm based on a convolutional neural network named YOLOv5s is studied.YOLO algorithm is the most classic among numerous target recognition and detection algorithms.Using YOLO algorithm can save a lot of selection work involving human participation,and also save a lot of tedious deep learning time.The reasonable use of YOLO algorithm can obtain a neural network model with very stable computing performance and high recognition accuracy.However,with the continuous optimization and improvement of YOLO algorithm,the deep learning of artificial intelligence will also greatly improve the accuracy of target recognition.The data generated by YOLO algorithm in the deep learning process will become larger and larger,which makes the original YOLO algorithm have a lot of inconvenience in the actual work of large-scale agricultural disease and pest detection.Based on the above analysis,this paper adopts the more advanced YOLOv5s algorithm to build the convolutional neural network model.Before the target recognition of agricultural diseases and pests,this study first constructed a disease and pest image data set specially used for training,including 20 kinds of greenhouse diseases and pests images.Then,the YOLOv5s algorithm was optimized and improved,and the algorithm was constructed based on GhostNet module.At the same time,deep convolution operation and attention mechanism were introduced.The neural network model constructed by the improved YOLOv5s algorithm can quickly detect disease and pest targets,and the generated network model has the characteristics of light volume,fast operation and strong portability.In order to further apply the research results to the actual agricultural production environment,a set of disease and pest image recognition system was developed based on Python language in this paper.The images of greenhouse diseases and pests were captured by cameras,and then the target recognition was carried out through the implanted YOLOv5s neural network model.Through the simulation experiment,the software can accurately complete the task of multi-target identification,and then complete the identification and detection of agricultural diseases and pests.