Research on adaptive inspection strategies and planning for unmanned aerial vehicles based on an improved ant colony optimisation algorithm

To achieve complementary advantages among different optical modal information and support power equipment fault detection and localization tasks, this paper uses visible-light images to enhance the texture information of infrared images. To address the difficulty of existing infrared–visible image registration techniques in accurately aligning local fine structures of power equipment, an adaptive registration algorithm with supervision and retraining (ARSR) is proposed for the first time. The algorithm mainly includes a dual-order anisotropic Gaussian directional derivative (Dual-AGDD) mechanism and a double-view matching parameter retraining (DVMPR) framework. First, Dual-AGDD is proposed to complete feature point screening and orientation. The first-order AGDD is used for adaptive local refined corner detection of power equipment, while the second-order AGDD constructs Gaussian feature triangles to determine the main directions of feature points, and a local intensity invariance method is adopted to construct feature descriptors. Then, the DVMPR framework is proposed to constrain and correct image perspective scale and field-of-view rotation. Finally, support vector regression is improved based on the 3σ principle to remove mismatched points and complete heterogeneous data registration. Experimental results show that, when registering heterogeneous images of power equipment under different rotation and scale differences as well as different environments, the proposed algorithm achieves an average localization error of 2.65 and an average registration precision of 98.57%. It demonstrates strong image rotation invariance, scale invariance, and environmental robustness, and significantly outperforms existing registration algorithms such as CAO-C2F and SuperPoint-SuperGlue, thereby improving the registration accuracy of heterogeneous images of fine structures in power equipment.