Precision pest management in agriculture using Inception V3 and EfficientNet B4: A deep learning approach for crop protection

Ensuring agricultural productivity and sustainability requires timely and accurate pest identification, as pest infestations significantly impact crop yield and food security. With increasing reliance on smart farming practices, artificial intelligence presents an effective solution for early pest detection. This study aims to evaluate and compare the performance of two state-of-the-art deep learning models, Inception V3 and EfficientNet B4, in identifying agricultural pests using transfer learning techniques. Both models were trained and tested on the IP102 dataset, which contains 102 distinct pest classes. The methodology involved leveraging advanced data preprocessing steps, including high-quality image selection and data augmentation, to improve model generalization. Transfer learning and fine-tuning were applied, with optimization of hyperparameters such as learning rate, batch size, and optimizer type to enhance model performance. Experimental results revealed that EfficientNet B4 significantly outperformed Inception V3, achieving a training accuracy of 96.32% and testing accuracy of 82.54%, compared to Inception V3′s 75.23% and 69.00%, respectively. The study also addressed class imbalance, further improving classification accuracy across varied pest types. These findings suggest that EfficientNet B4 is highly effective in detecting a wide range of pests and can be deployed in precision agriculture tools. The application of such AI-powered models holds the potential to revolutionize pest management by enabling early intervention and reducing crop loss. Also, the study contributes to several Sustainable Development Goals (SDGs): SDG 2 by boosting crop yields, SDG 12 by minimizing pesticide usage, SDG 13 by supporting climate-resilient farming, and SDG 15 by preserving biodiversity and encouraging eco-friendly practices.