Light-weight transmission tower angle steel embossed character recognition method based on deep learning

• Main Authors: YING SHENSHUN, FANG JIANHAI
• Format: Patent
• Language: Chinese; English
• Published: 07.05.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention discloses a light transmission tower angle steel embossed character recognition method based on deep learning, and belongs to the technical field of embossed character recognition. On the basis of a YOLOV5 reference model, a fusion type ShuffleNetV2-SE network architecture is provided, a strategy fusing a singular value decomposition (SVD) pooling layer and a random inactivation mechanism is designed, the ShuffleNetV2-SE network architecture is optimized so as to enhance the performance of the ShuffleNetV2-SE network architecture in detail recognition and noise suppression, and the ShuffleNetV2-SE network architecture has the advantages of being high in robustness and high in robustness. The flexibility and accuracy of the model in capturing, processing and reconstructing high-dimensional features are effectively improved; secondly, through the designed self-adaptive multi-dimensional feature coupling detection head, deep extraction of an input feature map and high reconstruction of spatial feat