基于图像识别的矿车装载物识别系统

Mine Vehicle Loading Recognition System Based on Image Recognition

目前我国矿山建设领域很多系统存在功能不完善、可操作性差、智能化程度较低、数据无法得到深度利用等问题。引入图像识别技术,对矿车装载物进行识别以此来提高分运效率,是解决上述问题的一个可靠途径。文章首先综合评估Faster R-CNN、SSD、YOLO、RetinaNet四种图像识别算法在矿车装载物分类中的实际性能,发现YOLO模型最佳。其次使用轻量级的MobileNet V3网络替换YOLO的特征提取网络,优化后的模型在保证精度的基础上,大小变为原有的1/5,且在不同环境下都能进行有效识别并分类。最后设计了人机交互界面并搭载触摸屏,实现了整套矿车装载物智能识别系统。

At present,many systems of mine construction field in China have problems of imperfect functions,poor operability,low degree of intelligence,and the data cannot be deeply used.The image recognition technology is introduced to identify the mine vehicle loading so as to improve the separation efficiency,which is a reliable way to solve the above problems.Firstly,this paper evaluates the actual performance of four types of image recognition algorithms such as Faster R-CNN,SSD,YOLO,and RetinaNet in vehicle loading classification comprehensively,and it is found that YOLO model is the most suitable.Secondly,the feature extraction network of YOLO is replaced with a lightweight MobileNet V3 network,and the size of optimized model becomes the 1/5 of the original size on the basis of the guaranteed accuracy.And it realizes effective identification and classification in different environments.Finally,it designs the human-computer interaction interface and touch screen is equipped,realizing a set of complete and intelligent recognition system for mining vehicle loading.