LPF-Induced Mixup for Green Pepper Segmentation in Hyperspectral Images

In agricultural production, the automatic segmentation of green pepper images is challenging due to the similar color between the peppers leaves, and stems. This study proposes a hyperspectral data augmentation technique that combines low-pass filtering (LPF) with Mixup, aiming to improve the accuracy of green pepper segmentation. We conducted two sets of experiments to evaluate the effectiveness of this method: 1) A binary classification of fruit and non-fruit pixels on a multilayer perceptron(MLP); 2) Six-fold cross-validated image segmentation on CNN. The experimental results on the test set showed that the proposed method improved the average F1 score by 15.9% in the classification task and 2.1% in the convolutional neural network compared to traditional methods. It also helped to enhance model stability and generalization ability.