Yanming Hui

In low-light conditions, the detection scene can be harsh, some fundamental image features of the target to be lost, which can result in the disappearance of essential visual characteristics of the object to be detected. They have failed to balance the connection between the low-level semantic information of low-light images and normal images. This article proposes an algorithm for weak-supervised and adaptive object detection in the low-light environment for YOLOV7 (WSA-YOLO) that utilizes adaptive enhancement to effectively improve object detection capability in low-light environments, addressing this practical issue. The proposed decomposition network decomposes the image into reflectance and illumination maps, which are then enhanced separately. The proposed adaptive residual feature block (ARFB) effectively utilizes the feature correlation between low-light and normal-light images and shares the weights between them to improve parameter reuse capability during parameter prediction using the parameter prediction block. The proposed adaptive adjustment block and consistency loss function are used together to enhance the brightness and suppress noise. Finally, the you only look once (YOLO) framework is utilized for object classification, regression, and prediction. Using the metric mean average precision (mAP) for evaluation on the recognized datasets, the proposed WSA-YOLO has a performance improvement of about 8% in peak signal-to-noise ratio (PSNR), structural similarity index, and natural image quality evaluator (NIQE). And the increase in mAP is about 9%.

Most LLIE algorithms focus solely on enhancing the brightness of the image and ignore the extraction of image details, leading to losing much of the information that reflects the semantics of the image, losing the edges, textures, and shape features, resulting in image distortion. In this paper, the DELLIE algorithm is proposed, an algorithmic framework with deep learning as the central premise that focuses on the extraction and fusion of image detail features. Unlike existing methods, basic enhancement preprocessing is performed first, and then the detail enhancement components are obtained by using the proposed detail component prediction model. Then, the V-channel is decomposed into a reflectance map and an illumination map by proposed decomposition network, where the enhancement component is used to enhance the reflectance map. Then, the S and H channels are nonlinearly constrained using an improved adaptive loss function, while the attention mechanism is introduced into the algorithm proposed in this paper. Finally, the three channels are fused to obtain the final enhancement effect. The experimental results show that, compared with the current mainstream LLIE algorithm, the DELLIE algorithm proposed in this paper can extract and recover the image detail information well while improving the luminance, and the PSNR, SSIM, and NIQE are optimized by 1.85%, 4.00%, and 2.43% on average on recognized datasets.

First, based on a given low-light image, moderately exposed and overexposed images are generated through the Exposure Image Prediction Model (EIPM).Then all three generate feature maps by a Feature Extraction Module (FEM),and then the feature maps of them are denoised. Subsequently, the feature map of the low-light image is fused with the feature maps of the two exposure rates images respectively. Finally, an adaptive weighted fusion method is used to obtain ultimate enhanced image. Experiments show that by comparing with the current SOTA algorithms,the proposed ELLIE algorithm, both in the detail feature retention and in color recovery have been significantly improved, and the subjective and objective evaluation indicators also have obvious advantages.