Strawberry Pests and Diseases Recognition with Self-Supervised Learning

This paper proposes a self-supervised learning approach to improve the recognition accuracy of strawberry crop pests and diseases, which can cause significant economic losses and reduce crop yields. Traditional methods for identifying pests and diseases are time-consuming and inefficient. Self-supervised learning can leverage unlabeled data to learn valuable representations without relying on manual annotations. The proposed two-stage approach involves self-supervised pre-training on the strawberry crop pests and diseases dataset and fine-tuning using a task-specific learning approach. Results show that this approach significantly outperforms the training from scratch and obtains similar results compared to the supervised pre-training approaches, with a highest recognition accuracy of 98.47% using ResNet-50 as the backbone model.