Parts Grasping by Intelligent Industrial Robots Based on Image Matching Algorithms and CNNs

Abstract

This study proposes an intelligent industrial robot part-grasping optimization model that integrates image-matching algorithms with convolutional neural networks. The model employs SuperPoint-based feature matching and introduces multi-scale convolution and a channel attention mechanism to enhance feature extraction. A generative convolutional neural network module further predicts the optimal grasping posture end-to-end. Experimental results show that the proposed image matching model achieves a matching accuracy of 97.8%, a repeatability of 92.5%, and a feature detection speed of 12.1ms per frame. The model maintains stable performance under complex lighting conditions and multi-view industrial environments, demonstrating strong robustness and real-time capability. These improvements enable more reliable perception and decision-making in robotic grasping tasks, thereby supporting flexible automation and the stable deployment of industrial robots in real manufacturing scenarios.