The advancement of image segmentation models has been significantly driven by the development of convolutional neural networks (CNNs). These homogeneous architectures demonstrate strong performance in feature extraction through the use of convolutional operations applied via a sliding window mechanism. However, their effectiveness tends to diminish when training data is limited, often resulting in overfitting. In this study, we propose a novel regularization method inspired by the electric force model for edge detection. A custom layer, referred to as the CVVEFM layer, is developed and integrated into a U-Net architecture for the segmentation of agricultural aerial imagery. To evaluate the effectiveness of the proposed method, comparisons were conducted against conventional regularization techniques, such as Dropout and ReLU. Experimental results demonstrate that the CVVEFM layer outperforms existing methods in terms of evaluation metrics: accuracy, precision, recall, and f1-score, as well as training stability. These findings highlight the potential of physics-inspired regularization approaches to enhance model generalization in data constrained environments.