Dynamic Encoding-Driven Cross-Channel Feature Cable Segmentation Network

In recent years, channel attention has been extensively explored in cable segmentation networks and has achieved significant accomplishments. However, most researchers still overlook the importance of cross-channel information. Due to the insufficient extraction of channel data, important details within the channels are often discarded, details that could enhance segmentation performance by providing intrinsic semantic information. To address this issue, we propose a dynamic encoding-driven cross-channel feature cable segmentation network. Specifically, we introduce a multidimensional pooling channel attention mechanism that integrates global statistical features, local prominent features, and the dispersion of features within channels. This allows for more refined and dynamic adjustment of channel information, enhances the perception of feature details, and improves the accuracy of cable information weights within channels. On challenging datasets, our method has achieved state-of-the-art segmentation performance.