Research On Chip Vision Counting Algorithm Based On Deep Learning

As the core component,the chip is widely used in various electronic products and industrial equipment,and has become an indispensable part of daily life and industrial production.The chips are assembled in the form of trays,and the number of chips in each tray needs to be counted before leaving the factory.The traditional manual counting method based on length or quality has many problems,such as complicated operation,slow speed and low counting precision,which can not meet the needs of production.In this paper,an intelligent chip visual counting algorithm based on deep learning is designed,which can effectively solve the current chip counting problem and meet the needs of actual production.The main work of this paper is as follows:(1)According to the chip counting requirements,the X-ray imaging scheme is adopted,and the overall flow of the chip counting algorithm based on deep learning is designed.(2)In view of the tedious operation of manual annotation with data set,a semi supervised automatic annotation algorithm based on adaptive threshold segmentation is designed,which cooperates with manual calibration to simplify the data annotation and improve the work efficiency.(3)Aiming at the problem that the training model takes too long,combined with a priori information,a sample selection strategy based on redundant elimination is designed to simplify the samples,which can ensure the accuracy of the model and accelerate the training speed.(4)The chip counting is abstracted as a target detection problem,and a chip counting algorithm based on the yolov3 network is designed.(5)The above algorithm in this article is realized as a annotation module,training module and test module through programming,and integrated into a set of chip counting software to realize automatic counting of the entire chip on the prototype.The actual sample was used to verify the designed algorithm in this paper.The experimental results show that the training time after redundancy elimination strategy is improved by 33%,the average accuracy of automatic counting is 98.6%,and the average time of each tray is 1.14 s,which basically meets the production requirements.