Document Type : Original Research
Authors
1
Biosystems Engineering Department, Faculty of Agriculture, Shahid Bahonar University of Kerman, Kerman, Iran.
2
University of California, Agriculture and Natural Resources, CA, United States
10.22103/bbr.2026.27265.1151
Abstract
Recognizing the challenges associated with conventional pest identification methods, which are labor-intensive, detrimental, and restrictive, the development of effective predictive models for accurate crop pest identification is paramount. This research utilized machine learning and deep learning techniques to extract features and classify tomato pests. Transfer learning was applied to a dataset of 1304 images of eight distinct tomato pests to expedite training. Critical pest characteristics were extracted employing convolutional neural network models, including AlexNet, GoogLeNet, ResNet101, and VGG16. Naïve Bayesian, Support Vector Machine, and K-Nearest Neighbors. With data augmentation and image preprocessing, ResNet101 achieved the highest performance, attaining 92.4% accuracy and surpassing the other models. The classification capabilities of ResNet101 and the SVM classifier were evaluated employing F1 scores across eight pest species (Bemisia argentifolii, Helicoverpa armigera, Myzus persicae, Spodoptera exigua, Spodoptera litura, Thrips palmi, Tetranychus, and Zeugodacus cucurbitae), yielding F1 scores of 69/74%, 92/87%, 88/88%, 75/69%, 52/65%, 76/9%, 35/35%, 72%, and 100%, respectively. The integration of machine learning and deep learning methodologies enables faster pest detection for experts and farmers, improves feature extraction, shortens training time, and ultimately increases crop yields and reduces economic losses.
Keywords
)