<p>This paper presents a novel approach to vehicle small target detection in aerial images, referred to as RCV-Fusion Net. Intelligent transportation systems critically depend on precise vehicle detection to improve traffic efficiency and safety. However, multimodal vehicle small target detection faces several challenges, including insufficient feature extraction, inadequate multimodal information fusion, and problems related to misdetection and omission. To address these challenges, we propose an algorithm that combines dynamic feature re-extraction with cross-modal feature fusion. Our dynamic feature re-extraction module employs a two-branch strategy, utilizing Efficient-LSTM for global feature extraction and Adaptive Channel CNN for local feature extraction. A dynamic weight generation module is introduced to balance the contributions of these features, effectively suppressing irrelevant background information. Furthermore, a cross-modal fusion module is designed to maximize the integration of visible and infrared modal information, thereby exploiting their complementarity. To enhance training efficiency, depthwise separable convolution is introduced, and the Neck part of the network is redesigned. Experimental results on the VEDAI and Drone Vehicle datasets demonstrate that RCV-Fusion Net significantly outperforms existing methods, achieving mean average precision (mAP) scores of 87.2% and 86.5% on the respective datasets, thereby validating its effectiveness for detecting small vehicle targets.</p>