Research on Plant Pest and Disease Classification Based on Self-Supervised Learning

In everyday life, the classification of plant pests and diseases in images is deemed crucial. However, the collection and manual labeling of databases containing images of plant pests and diseases are considered time-consuming tasks. In this study aims to address this challenge by extracting image features from extensive, unlabeled datasets of plant pests and diseases. To achieve this, a self-supervised learning method utilizing the momentum contrast (MoCo) technique for image classification is introduced. Specifically, an approach named modify_arg_MoCo is proposed, which includes appropriate data augmentation techniques and learning rate settings based on a cosine schedule. Additionally, the impact of different momentum values, the presence of an MLP layer, and epoch parameters on classification outcomes was investigated. Experimental findings indicate that the modify_arg_MoCo method can achieve good classification accuracy (94.31%) in plant pest and disease images.