<p>Low-light images often require enhancement techniques to improve visibility, particularly owing to differences in RGB channels. This study introduces the Distended Pixel Replacement Technique (DPRT), which employs a Partial Convolution Network (PCN), an adaptation of traditional convolutional neural networks with a dual training process. In the first training phase, the network detects tampered pixels after identifying the distended pixels, and in the second phase, it reverts these tampered pixels to their original states. These partial training phases work in tandem to detect concatenated pixels, guiding the replacement of light or dark pixels based on the recognized features. Additionally, the unique characteristics of each RGB channel were analyzed to optimize contrast enhancement. The proposed method improves pixel detection and replacement, leading to a significant visibility enhancement in low-light images. The DPRT with PCN was evaluated using accuracy, precision, sensitivity, specificity, and false modification rates, achieving 13.33% higher accuracy, 12.64% higher precision, and 10.74% fewer false modifications than existing techniques. The method can be fully implemented using MATLAB with deep-learning toolboxes.</p>