Deep Learning-Based Approach for Agricultural Pest Recognition in Smart Farming Applications
Abstract
Pest infestation in agriculture is a significant challenge to the world food security and production. This research uses artificial intelligence to create a deep learning pest recognition system to improve accuracy in agriculture through the implementation of pest recognition automation. They trained and tested several models using a dataset of 5,494 images that comprised 12 classes of pests, which included VGG16, ResNet-50, MobileNetV3, Inception-v3, as well as a modified CNN architecture. The models were evaluated using classification accuracy, precision, recall, and $F_{1}$-score. The CNN brought about customization showed competition, whereas attention mechanisms and transfer learning enhanced feature extraction and interpretability. Although the accuracy is high, there are still difficulties related to the limitation of datasets, computational power of light devices, and the inability to deploy it in the remote farms. On the whole, this piece of writing demonstrates how optimized deep learning models could help to facilitate proper pest management, improve crop yields, and embrace smart farming sustainability efforts worldwide.