基于改进视觉RWKV的红外图像小目标检测算法OA

Infrared small target detection has the advantages of low cost,non-contact sensing,and passive imaging,but it still faces signif-icant challenges under complex background conditions,such as small target size,low contrast,and strong background interference.To ad-dress the limited global modeling capability of convolutional neural network-based methods and the high computational complexity of Transformer-based approaches,we explore the introduction of the RWKV architecture into infrared small target detection and propose an improved visual RWKV-based detection algorithm,termed RWKV-ISTD.The proposed network adopts a dual-branch structure to sepa-rately extract local and global features.An improved visual RWKV auxiliary encoder is employed to model global contextual informa-tion,while a residual convolution-based main encoder is used to effectively capture local detail features.Considering the characteristics of infrared small target images,a novel multi-scale shift module combined with wavelet transform is designed to enhance multi-frequency feature representation during global feature extraction.In addition,an improved feature fusion module is introduced to enable deep inter-action between local and global features.Extensive experiments conducted on the NUAA-SIRST and IRSTD-1K public datasets demon-strate that the proposed method effectively improves infrared small target detection performance under complex backgrounds,while exhibi-ting advantages in model complexity control.These results indicate that the proposed approach achieves a reasonable balance between de-tection accuracy and computational efficiency,providing a feasible solution for infrared small target detection tasks.