Research On Detection Technology Of Engineering Vehicles Under Warehouse Scene Based On Lightweight Network Architecture

The transportation and storage of raw materials is an important part of the operation of coal mining enterprises.The dust generated during the loading and unloading of raw materials by vehicles in the warehouse not only seriously pollutes the air,but also is an important issue affecting the safety of the warehouse.Therefore,the detection of engineering vehicles in the storage scene has also become an important research topic in the production safety supervision process of coal mine enterprises.Traditional warehouses generally use a combination of manual inspection and instrument inspection to control engineering vehicles.Safety supervision personnel analyze the progress and status of on-site vehicle loading and unloading materials by observing monitoring images.At the same time,safety supervision personnel will also combine on-site air quality monitoring equipment to schedule on-site loading and unloading progress.This method not only wastes manpower,material resources and financial resources,but also makes it difficult to ensure accuracy and real-time performance for monitoring large areas.Based on the actual situation of the site,this paper designs a detection and positioning system for engineering vehicles that can be used in storage scenarios,which can realize the detection of engineering vehicles in the area by using the object detection network in the process of loading and unloading materials of engineering vehicles.The system can further determine the movement state of the object and locate the location of the engineering vehicle that has parked for loading and unloading materials for a long time in the area.Finally,the system can provide object position coordinates for the on-site water spray cannon controlling server.The water spray cannon controlling server can spray water mist to accurate coordinates to solve the problem of environmental pollution and danger of dust explosion caused by dust generated by the loading and unloading of raw materials of engineering vehicles in the storage scene.This subject mainly studies the object detection technology of engineering vehicles in the intelligent detection system of warehousing,which is mainly for bulldozers and trucks in the warehousing scene.First,build a monitoring system on site and use the front-end images collected by the monitoring camera to make a dataset of engineering vehicles.Annotate the filtered effective samples to obtain a complex storage scene engineering vehicle dataset.Secondly,in view of the complex situation in the warehouse scene and the coexistence of multiple scale targets,an improved engineering vehicle detection network based on lightweight network architecture is designed.The network integrates the Spilt-attention and modifies the architecture of the network branch parts.This paper replaces the activation function of the convolutional layer in the two multi-scale branches in the network with Dynamic Re LU and uses the self-built dataset to train the improved network.Then,this paper obtains the multi-type engineering vehicle detection model in the warehouse scene.At last,a set of engineering vehicle detection and positioning system oriented to storage scenarios is developed.In this paper,the improved network is tested on the self-built engineering vehicle dataset and the COCO dataset.The improved network is compared with a variety of current detection networks in these datasets.The experimental results show that,in terms of accuracy,the improved network designed in this paper can achieve AP50 of 43.0 on the COCO dataset and m AP of 98.2 on the self-built engineering vehicle dataset.In terms of detection speed,our method can reach a speed of 154 FPS with one 2080 Ti GPU,which meets actual application requirements and can be well applied to image detection of multiple types of engineering vehicles in storage scenarios.