An object detection network for wear debris recognition in ferrography images

The intelligent recognition of wear debris in ferrography images is a great challenge. Aiming at the detection characteristics of multi-scale objects, small objects and overlapping objects in ferrography images, a real-time on-line object detection model combining feature fusion, self-attention mechanism and improved NMS is proposed. Based on YOLOv3, a fully convolutional network based on feature fusion is constructed, and a residual unit with self-attention mechanism and an object selection mechanism based on improved non-maximum suppression are introduced. The performance of the model is compared with three advanced object detection algorithms on a great number of ferrography images. The experimental results show that the model has high detection accuracy, short detection time and strong adaptability. The proposed model can be further developed and applied to fault diagnosis and condition monitoring of machinery and equipment in future.