Efficiency and Scalability of EfficientNetV2-L for Large-Scale Pest Classification

Reducing damage to buildings, plants, and crops depends on efficient pest control in urban, forestry, and agricultural settings. Many times labour-intensive and prone to mistakes, traditional pest detection techniques are Using current developments in machine learning and image processing technology, this paper investigates the use of Convolutional Neural Networks (CNNs) to automate and improve pest classification. This work uses the EfficientNetV2-L model to obtain a strong classification framework using a large dataset of pest images classified into twelve different classes: ants, bees, beetles, caterpillars, earthworms, earwigs, grasshoppers, moths, slugs, snails, wasps, and weevils. To solve over fitting and maximise model performance, the suggested approach consists in data augmentation, a bespoke classification head, and dropout layers. Results show an overall accuracy of 88%, high precision, recall, and F1 scores for many pest categories. Though some categories like slugs and snails offer chances for additional improvement, the model shines in separating between several pest species. This work emphasises the great potential of CNN-based methods in providing improved accuracy and efficiency as well as in addressing pest detection and control. This work opens the path for treatments that are more exact, better agricultural results and environmentally sustainable practices by including automated classification into pest management plans.