<p>The contrast and saturation of images captured under adverse conditions can differ significantly from normal images. These variations lead to a domain shift, which reduces the effectiveness of object detection methods applied to normal datasets under such conditions. Existing methods often ignore the effects of non-uniform contrast changes in adverse environments. Relevant research indicates that humans’ inherent illumination invariance during image feature extraction can mitigate the effects of contrast changes. Inspired by this, we proposed a bio-inspired primary visual cortex enhancement network, called BPVENet. Specifically, drawing from the antagonistic mechanisms of simple cells in the primary visual cortex, we have designed a directed difference convolution. This technique leverages weight difference and adaptive directional modulation of convolution kernels to extract locally contrast-invariant contour features in images with non-uniform contrast variations. By simulating the information processing mechanisms of the primary visual cortex of the human brain, we designed an antagonistic enhancement network that can alleviate performance degradation caused by domain shifts in images by utilizing illumination invariance. BPVENet achieved mAPs of 52.63 and 54.93 on ExDark and RTTS, respectively. Compared to 9 methods, including DE-Yolo and BAD-Net, BPVENet shows a distinct advantage in helping object detection algorithms adapt to various adverse conditions.</p>