IMAGE RECOGNITION TECHNOLOGY BASED ON DEEP LEARNING IN AUTOMATION CONTROL SYSTEMS

In order to solve the problem of recognizing multiple product images, the author proposes the research of image recognition technology based on deep learning in automated control systems. Firstly, the FasterRCNN method is improved by proposing a non class specific FasterRCNN, which can be used for pre annotation of product images by training only on publicly available datasets. Due to the use of position correction networks, the pre annotation effect is more accurate than that of candidate region networks. Then, combining Grab cut with non class specific FasterRCNN, a sample enhancement method was proposed to synthesize a large number of training images containing multiple products and use them for model training. In addition, based on non class specific FasterRCNN, a re identification layer was proposed to improve detection accuracy. In the end, the recognition and positioning of multiple products achieved a recall rate of 93.8% and an accuracy of 96.3%.