YOLO-CAKM: Integrated multi-mechanism intelligent scrap classification and unmanned quality verification

As the only alternative raw material of iron ore in steelmaking process, scrap steel is of great significance to realize the sustainable development of iron and steel industry. However, the accurate identification of scrap grade has always been a big challenge faced by iron and steel enterprises. The traditional manual classification method has some problems such as strong subjectivity, low efficiency and difficult to guarantee impartiality. To solve this problem, In this paper, we propose a kind of non-human quality checking method for scrap classification based on machine vision, YOLO-CAKM (Coordinate Attention + Arbitrary Kernel Convolution + Multi-Scale Cross-Axis) Module). The method combines Coordinate Attention, Arbitrary Kernel Convolution, and Multi-Scale Cross-Axis Module. The spatial structure cognition ability, morphological adaptability and feature extraction accuracy of scrap images were significantly improved. Specifically, the coordinate attention mechanism enhances the spatial localization ability of the model through direction perception and position sensitive attention force. The changeable nuclear convolution technology ADAPTS flexibly to the changeable forms of scrap steel. Multi-scale cross-axis module mining multi-scale features and global background information, highlighting key features and reducing the interference of irrelevant features. The experimental results show that the YOLO-CAKM model performs well in the scrap classification task, with an accuracy of 89.96%, which is significantly better than the existing mainstream models. This method not only meets the actual production needs in terms of accuracy, real-time and identification efficiency, but also provides strong technical support for the automation and intelligence in the scrap procurement process of iron and steel enterprises, and promotes the intelligent development of the iron and steel industry.