A High-Performance Pest Detection and Classification Model Using Pyramid U-Net Fusion Network (PUFNet) and Partial Reinforcement Optimizer (PaFO) for Precision Farming

Pest identification and categorization in vegetable crops are essential to support high agricultural yields and food security. Control of pests can be achieved if the pests are identified at an early stage in the crop’s development, thus enabling eradication or at least reducing the yield loss and harming some of the yield quality, reducing the use of toxic chemicals and pesticides. Existing pest detection models present several issues, including low accuracy, inability to apply to a wide range of pest kinds, and the need for significant computational resources. These tenders often lead to missed detections and increased numbers of false positives or negatives; pest control is then not effective. To resolve these problems, we introduce the Pyramid U-Net Fusion Network (PUFNet), which is novel and better for pest detection and classification. In the design of PUFNet, the pyramid structure is combined with U-Net so that multi-scale features are utilized to enhance the fusion of related content. Further, we propose the Partial Reinforcement Optimizer (PaFO) for the tuning of parameters, which employs P-R learning to improve the existing model performance. The proposed PUFNet performs better than the existing models of pest detection in all key metrics. It achieves an accuracy of 98.5% and a precision of 98.4%, much better than models like CNN (95%, 84%) and RNN (97%, 96.8%). In addition, PUFNet achieves a recall of 98.45% and an F1-score of 99%, much better than CNN and RNN.