Aerial Infrared Target Recognition Algorithm Based on Multi-Feature Fusion

During the process of aerial infrared target recognition, the algorithm's performance is degraded by the interference of large area masking targets and the multi-scale changes in target shape. To address these challenges, a multi-feature fusion-based aerial infrared target recognition algorithm is proposed. Firstly, to mitigate the variations in infrared target features with changing scales, the HOG features of infrared images are extracted and fused with depth features. Secondly, a multi-scale hybrid dilated pyramid structure is devised to capture multi-scale global fusion features. Subsequently, an adaptive feature fusion mechanism is employed to dynamically enhance the multi-scale global fusion features and HOG features, which are then fused to obtain hybrid depth features. Finally, tests conducted on extensive datasets demonstrate that the algorithm achieves an average recognition accuracy 3% higher than that of the GoogLeNet algorithm, thus validating the effectiveness of the proposed algorithm.