Authors: Dr.B.Sai Venkata Krishna, G. Kavitha
Abstract: One of the main industries promoting both global food security and economic expansion is agriculture. However, because they significantly lower crop output and quality, insect infestations remain a danger to agricultural productivity. As a result, accurate insect pest identification is crucial to avoiding needless pesticide use and putting prompt, efficient pest control measures into place. Conventional pest detection methods are labor-intensive, time-consuming, and less successful for extensive agricultural monitoring because they mostly rely on manual inspection and handcrafted feature extraction. Recent developments in deep learning have made it possible for automated image analysis systems to accurately identify intricate visual patterns. Convolutional neural networks (CNNs) are used in this study's deep learning-based architecture for autonomous agricultural pest classification. First, several methods for identifying pests are analyzed to determine their benefits and drawbacks. The suggested approach uses CNN architectures and transfer learning to categorize insect pests gathered from various agricultural datasets. The classification model is trained and assessed using three benchmark datasets: Xie1, Xie2, and NBAIR. The Xie1 and Xie2 databases have 24 and 40 insect categories, respectively, whereas the NBAIR collection has photos of 40 insect classes gathered from various crops. In-depth experiments assess classification performance by contrasting the suggested CNN model with numerous well-known deep learning architectures. According to experimental results, the suggested framework maintains remarkable precision, recall, and F1-score across all datasets while achieving high classification accuracy. By improving precision agricultural decision-making, reducing crop losses, and early pest detection, the study shows how CNN-based image classification is useful for intelligent pest identification and how it may help sustainable agriculture.
