Early Detection and Prediction of Pests in Field Crops Using Transfer Learning

Abstract

Listen

Translate article

This research study addresses the problem of early detection and prediction of pests in field crops. The primary objective of this research is to identify and distinguish pest species from an open-source dataset that includes 5,494 images across 12 classes. We developed an efficient model with a high probability of detecting pests in field crops using pre-trained models such as EfficientNetV2 and deep learning techniques. We applied hyperparameter tuning to the model to enhance its accuracy. Our proposed model is designed to detect and predict pests at an early stage, thereby preventing crop damage. Experimental results demonstrate that the performance of the proposed model is more accurate and precise compared to state-of-the-art existing studies. The F1 scores of the model for different classes of pest images are as follows: Ants 0.96, Bees 0.98, Beetles 0.97, Caterpillars 0.98, Earthworms 0.95, Earwigs 0.97, Grasshoppers 0.96, Moths 0.96, Slugs 0.97, Snails 0.99, Wasps 0.99, and Weevils 0.98. The overall accuracy of the model across all classes is 97.17. These results demonstrate the improved performance of the proposed model for early pest detection. In the agricultural sector, this model can be immensely beneficial, aiding in quick, accurate, and reliable pest detection to support decision-making processes. Identification of pest occurrence at their early stages leads to actions on interventions, which helps in reducing crop losses avoids unnecessary spraying for chemicals, and ensures sustainable eco-friendly agricultural practices. An approach like this would help in maintaining food security and economic sustainability of farmer communities.