基于红外视频与音频联合的无人机目标检测OA

To addressing the challenges of detecting small drones,such as weak targets,complex backgrounds,and potential confusion,a detection method based on multimodal information fusion is proposed.To enhance the accuracy and robustness of target detection,the proposed method employs infrared video and audio for initial target detection,subsequently obtaining the final detection result through decision-level fusion.For infrared video,the"tracking-then-detecting"approach is adopted,incorporating a dynamic saliency difference enhancement module.The module integrates gradient-grayscale features and motion information through a multimodal feature fusion mechanism,combined with a window scaling strategy guided by local entropy and time-domain motion verification,thereby enhancing the contrast between weak and small targets and the background.Additionally,a space-time trajectory coding and correlation module is introduced,utilizing an LSTM network for short-term trajectory feature extraction and trajectory-measurement matching degree calculation.The dynamic space-time fusion factor optimizes the data association process,addressing issues such as target occlusion and trajectory discontinuity.Regarding audio processing,the Mel spectrum is extracted and converted into logarithmic Mel spectral features,with a CNN model employed for acoustic feature recognition.Experimental results demonstrate that the proposed method outperforms existing methods in terms of accuracy(89.5%),recall(85.7%),and average precision(75.4%),providing an effective and feasible approach for drone target detection in complex environments.