A digital gauge reading algorithm based on an improved YOLOv8

An automatic reading algorithm based on improved YOLOv8 is proposed for digital instrument reading. It can be applied in scenarios such as the petrochemical port, where it is difficult to capture high-resolution images. To address the issue of limited feature learning for low-resolution images, the SPD-Conv module is introduced. By transforming the feature map from width to channel depth, the network can learn deeper features, helping to capture complex patterns and structures in images and improving detection performance for low-resolution pictures. Additionally, the channel and spatial-based attention mechanism CBAM is introduced on the foundation of SPD-Conv. It calculates the importance of each channel, focusing on more essential channel features and spatial features, enhancing the model's feature extraction capability. Wise-IoU is introduced to achieve dynamic focus for bounding box regression loss, improving the model's ability to locate numerical digits. Experimental results show that the improved YOLOv8 algorithm increases the model's mAP from 94.2% to 95.6% compared to the original model. The enhanced model exhibits improved generalization and detection efficiency for low-resolution images.