Research On Intrusion Detection Algorithm Of Port Dangerous Area Based On Deep Learning

In recent years,with the vigorous development of the port industry,the work of the port dock is becoming more and more frequent.However,it is accompanied by how the port guarantees the safety of the operators during the operation,and prohibits irrelevant personnel from entering the dangerous area during the operation.In order to solve the problem of dangerous zone intrusion in port operation,this paper proposes a dangerous zone intrusion detection algorithm based on deep learning,and builds the port dangerous zone intrusion detection system based on this algorithm.Since the detection task in this paper needs to take the triangular cone as the anchor point to circle the dangerous area to be detected,and the installation position of the field camera is basically on the mechanical arm of construction machinery,and the distance between the camera and the object to be detected is relatively long,the algorithm in this paper is the extended application of the object detection technology in the small eye detection task.The main work and research contents of this paper are as follows:(1)A port intrusion detection algorithm based on YOLOv4 is proposed.Considering the poor detection effect of the YOLOv4 network model on small objects and the low regression rate of network model detection,the transformer module is first introduced based on the original YOLOv4 network to improve the regression rate of the network model and increase the detection accuracy of the model.Secondly,this paper reconstructs the feature fusion module of the original network,adding subsampling connection and skip connection,so that the small object information in the input image can be more fully utilized,and the detection accuracy of the network model on small objects can be improved.The experimental results show that compared with the original network model,the performance of the improved network model is significantly improved,especially in the detection accuracy of small objects.The improved model preliminarily meets the requirements for detecting the intrusion of dangerous areas in ports under operational conditions and solves the problem of irrelevant personnel intruding into dangerous areas to a certain extent.(2)Aiming at how to speed up the model detection speed and reduce the number of parameters of the model for field deployment while ensuring high detection accuracy,a lightweight YOLOv4 algorithm based on a network model is proposed.Firstly,the feature extraction network of the original model was replaced with a lighter shufflenentv2.Then,according to the idea of obtaining as many feature graphs as possible with fewer operations in Ghost Net,a new CB structure is constructed.This structure can perfectly replace the convolutional module in the network so that the overall parameter number can be reduced while the detection accuracy of the model is almost constant,and the architecture design of the overall network model can be realized.(3)Make intrusion detection data sets of port dangerous areas.In view of the lack of a picture database of the port and engineering equipment during operation,a port data set containing 3800 pictures was independently made.The data set includes all kinds of complex scenes(such as fog,night,dust,strong light and other harsh conditions),and multi-scale field pictures of various engineering instruments,greatly enhancing the robustness of the algorithm.(4)Human-computer interaction page development.In view of the problem that the current object detection network is mainly carried out under the command window,in order to facilitate the field deployment,a human-computer interaction page based on MFC is designed and developed independently,which realizes the visual operation of the field object detection and the judgment of the intrusion detection behavior,and improves the intelligent level of port operation.