Drilling rod target segmentation recognition based on an improved GAN

Abstract

During the process of gas hole drilling, automatic loading and unloading drilling rod by robotic arm ensures the safety of personnel and drilling efficiency. Accurate recognition of drilling rod target is a prerequisite for precise positioning. However, the presence of dark and dust underground coal mines presents the great challenge in detecting and recognizing drilling rods during the automatic drill loading and uploading process. To solve this problem, We have designed a drilling rod target detection and segmentation technology based on generating adversarial network(GAN). Furthermore, we carried out experiments to compare the recognition performance of drilling rods of different colors, including black, blue, and yellow, in the dark and dusty environment. The results indicate that the drilling rod recognition method proposed in this paper demonstrates high accuracy and robustness even in dark and dusty environment, better than other commonly used segmentation networks. Notably, the recognition accuracy of yellow drilling rods surpasses that of blue and black drilling rods.