Fast Recognition and Location Method of Parts for Assembly Robot Based on Deep Learning Network
Aiming at the problems of low components recognition rate and poor robustness under complex conditions such as adhesion, stacking and light source interference, a fast recognition and position method of parts based on deep convolution neural network was proposed. A multilayer convolution neural network for real-time detection of parts was constructed. Firstly, multi-scale extraction of candidate regions for target images was carried out. Then,the feature vectors of candidate candidate regions were automatically extracted by convolution neural network. The softmax classifier was used to recognize and locate. The experimental results show that the method realized the recognition and location for multiple parts in complex environment. The accuracy rate is over 95.2%.The detection took only 78ms.The accuracy rate and real-time performance of the method proved its feasibility.
Paper
Full text
Fast Recognition and Location Method of Parts for Assembly Robot Based on Deep Learning Network
Semantic Scholar · Computer Science · 2019
Abstract
Aiming at the problems of low components recognition rate and poor robustness under complex conditions such as adhesion, stacking and light source interference, a fast recognition and position method of parts based on deep convolution neural network was proposed. A multilayer convolution neural network for real-time detection of parts was constructed. Firstly, multi-scale extraction of candidate regions for target images was carried out. Then,the feature vectors of candidate candidate regions were automatically extracted by convolution neural network. The softmax classifier was used to recognize and locate. The experimental results show that the method realized the recognition and location for multiple parts in complex environment. The accuracy rate is over 95.2%.The detection took only 78ms.The accuracy rate and real-time performance of the method proved its feasibility.