Structures Detection Based on CLSK Model Combined With Shadow Information Using High-Resolution Remote Sensing Images

Abstract

In light of rapid economic and urban growth, the proliferation of structures including transmission towers, signal poles, and wind generators has become evident. Consequently, precise object detection of these structures emerges as a pivotal approach to enhance infrastructure management. This technique establishes a robust foundation for achieving elevated efficiency, precision, optimized energy management, and heightened safety monitoring. This article introduces a novel model for detecting structure based on channel and large-scale selective kernel (CLSK) model using high-resolution images. The method is rooted in a two-stage target detection network, enabling simultaneous identification of both primary structures and their associated shadows. The incorporation of deformable convolutions augments the model's ability to extract intricate features. Moreover, the introduction of the innovative LSK attention module, along with the CLSK attention module, enhances the optimization of features gleaned from the network's core architecture. Simultaneously, the complete intersection over union loss function refines the network's focus by considering parameters such as center point distance and aspect ratio in addition to the conventional overlapping area. This comprehensive approach facilitates improved feedback on detection outcomes. Empirical evaluation of the proposed network underscores its superior performance when juxtaposed with both conventional network models and the rotating detection box network.

Keywords

• Attention mechanism
• complete intersection over union (CIoU)
• deformable convolution
• shadows
• structure