Research On Grain Silo Operation Safety Early Warning System Based On Intelligent Video Analysis

At present,the field of artificial intelligence is developing rapidly,and the application of computer vision technology in various fields of life and industrial production has been popularized.How to apply deep learning technology to the field of intelligent monitoring has become a key direction of research.Among them,human behavior recognition is a key research direction in the field of computer vision.It can not only develop in the fields of medical treatment and automatic driving,but also replace human beings to intelligently identify pedestrian movements in videos and provide intelligent solutions for improving the safety of public places.Therefore,how to improve the performance of the human behavior recognition algorithm in various scenarios is a hot issue in the field of video-based intelligent recognition.This paper proposes a time-domain segmentation network model based on the fusion of fast and slow neighborhood features(TSF)based on a dual-stream network architecture to address the problems of existing deep behavior recognition models that are prone to overfitting in training,high computational resource requirements,and inadequate extraction of long timedomain features,and uses the improved model to perform training tuning for unsafe behavior in a specific site of grain bin operation,and designs and implements an early warning system for safe grain bin operation based on intelligent video analysis.Specifically,this article has completed the following work:Firstly,to be able to improve the two-stream network structure for long-time action recognition in video recognition,an action recognition method using a neighbourhood fusion strategy to extract long time-domain features is proposed in combination with Slowfast networks for improving this model of TSN.First,the video is segmented and each segment is asymmetrically sparsely sampled to extract stage-spatial features through the Slowfast network.Then the feature matrices of neighbouring time domains are stitched together to obtain the phase features,and as the features are enriched the network gradually focuses on the action regions.Finally,the video-level results are obtained by averaging and weighting the stage results obtained at each stage.The experiments show that the improved method used in this paper allows the network to reach the optimal weights faster while effectively improving the recognition accuracy.Then,the unsafe working behaviour of grain silos is analysed,model training is tuned and applied,unsafe behaviour criteria are combined to classify grain silo working behaviour and generate data sets,while migration learning and hierarchical training methods are used to better train the model.Finally,an intelligent video analysis-based grain silo safety warning system is designed and implemented to automatically detect unsafe behaviours in grain silos,provide timely warning of unsafe behaviours to reduce accidents and improve safety management.