An Efficient and Accurate Object Detection Algorithm And Its Application

Abstract

In order to solve the problem that the recognition accuracy and recognition speed of the current deep learning target detection algorithm are not compatible, an efficient and accurate object detection algorithm was proposed and applied to the inspection of small-sized body stamping parts of an auto parts enterprise. Firstly, image data sets of 10 small-sized automobile body stampings were produced. The transfer learning method was used to train YOLO V3, YOLO V3-tiny and the model proposed in this paper. Then, three models were used for small-sized body stamping parts detection and recognition experiment. The experimental results show that our model can detect 37 images per second in the same sample and test environment, which is 208.33% higher than the YOLO V3 model; the average detection precision is 96.50%, compared with The YOLO V3-tiny model is improved by 4.40%, and the average detection precision is improved up to 22.58% on a single smaller size object. This study can provide visual navigation support for the small-size body stamping robot automatic sorting system.