Research On The Grab Detection Algorithm Of Portal Crane Based On Deep Learning

In recent years,with the vigorous development of the port industry,port throughput has continued to increase,and the demand for loading and unloading dry bulk cargo is also increasing at the port terminals.As we know,portal crane is one of the main handling tools of dry bulk cargo in the port,so the operation control of the grab is the key to improve the efficiency and intelligent degree of loading and unloading dry bulk cargo.In order to solve the problems of low efficiency and safety caused by the inability of human eyes to accurately determine the position of the grab during the operation of loading and unloading of dry bulk with the grab of portal crane,this paper proposes the grab detection algorithm of portal crane based on deep learning.The main work and research are as follows:(1)This paper designs a kind of grab detection algorithm of portal crane which is improved based on convolutional neural network of YOLOv3-tiny.In view of the fast speed of detection but low accuracy of the YOLOv3-tiny network model,firstly,the spatial pyramid pooling is introduced on the basis of the original network to extract the multi-directional features of grab;Secondly,depthwise separable convolution is introduced and the network width is expanded to form an inverted residual group,which is added to the network structure to deepen the network and improve the performance of detection,while maintaining a small number of parameters and reducing computational consumption;Finally,the feature fusion is performed by adding two dilated convolutional layers between the lower layer of the network and the upper layer to ensure that the receptive field of the fused feature map is expanded without loss of resolution.The experimental results show that the performance of the improved network model is significantly improved compared with the original model,which meets the real-time and accuracy of the grab detection in general condition of working.To a certain extent,it solves the problems of low efficiency and security.(2)In view of the low visibility of the grab caused by factors such as high concentration of coal dust,strong sunlight,foggy weather and other factors in complex scenarios,the grab detection algorithm based on Efficient Net and feature fusion is proposed in this paper.Firstly,the backbone network of YOLOv3-tiny is replaced by Efficient Net.Then the partial residual connection and the attention mechanism module are designed in front of the detection layer.Finally,instead of the traditional feature fusion method of FPN,the two-way feature pyramid structure based on M-Attention and PRN is proposed in this paper,so that the more details and position information of the grab at the bottom layer are fused with the strong semantic information of the high layer to make predictions,and the model design of overall network is realized.After experimental verification,the improved network model effectively improves the ability of grab detection.It solves the problem of false detection and missed detection of the grab in complex scenarios to a certain extent,which enhances the robustness of the algorithm.At the same time,the intelligent level of grab operation is improved by encapsulating and calling the model of optimal algorithm.