For fault-controlled channel sandbody reservoirs, where small faults are prevalent, sandbodies are thin, and the data have a relatively low signal-to-noise ratio, we propose an improved 3D ResU-Net++ network that integrates an enhanced DenseASPP module, CBAM based attention optimization, and depthwise separable convolutions to address the identification of low-order faults in complex settings. First, a DCNv2-enhanced DenseASPP module is introduced into the ResU-Net++ backbone. By densely connecting multi-scale DCNv2 convolution branches, the network's receptive field is markedly expanded, improving multi-scale contextual awareness of fine-grained faults and thin-layer sandbodies. Second, CBAM recalibrates channel and spatial attention weights to further highlight fault textures and structural boundaries in addition, standard convolutions are replaced with depthwise separable convolutions to boost the model's resolving power and computational efficiency. Finally, transfer learning is employed to enhance generalization. Applied to the Tong-61 fault block, the method achieved segmentation metrics of IoU and Dice up to 85.6% and 92.2%.Well-seismic comparisons and case studies indicate that the proposed approach can accurately delineate complex fault structures and finely depict the distribution of thin sandbody reservoirs, providing an efficient and effective technique for integrated structural-lithologic interpretation and high-resolution hydrocarbon exploration and development.