DESIGN OF INTELLIGENT VIDEO SURVEILLANCE SYSTEM FOR INTELLIGENT ORCHARD BASED ON DEEP LEARNING NETWORK

Aiming at the problem of low intelligence level of orchard management in China, this study develops an open intelligent video surveillance system for intelligent orchards based on deep learning target monitoring and behavior recognition algorithms. Based on the open-source distributed framework gRPC, the system improves the main network structure of YOLOv5 and optimizes the abnormal behavior recognition algorithm based on human skeleton. It not only realizes the dynamic allocation of servers and the number of cameras for multi-platform and multi-host communication and maximum utilization of existing computing resources, but also improves the recognition rate of small and partially occluded targets in multi-channel video and reduces the amount of network computations while ensuring the real-time monitoring accuracy of abnormal behavior. The test results show that the monitoring system’s accuracy is higher, the illegal operation monitoring accuracy is over 91.25%, and the abnormal behavior monitoring accuracy is above 61.78%. The surveillance system can meet the monitoring needs of intelligent orchards, reduce the cost of orchard management, and realize the scale and intelligent management of orchards, which is of great significance to promote the upgrading of orchard industry. © 2023, Politechnica University of Bucharest. All rights reserved.