Detecting and Classifying Household Insects in Iraq by Using Transfer Learning Models

Abstract

Household insects are a part of everyday life, yet they may be a menace in some different situations. Developing more efficient detection and classification methods is necessary since the conventional approach may be laborious and error-prone. By harnessing the ability of algorithms to handle enormous volumes of data and extract essential information, machine learning has emerged as a potential method for insect detection and classification. In this study, insect detection and classification were proposed based on machine learning and deep learning techniques. Transfer learning was used for feature extraction, a YOLOv8 for detection, and a support vector machine algorithm for classification. Transferring learning models may significantly boost detection and classification precision when learning about insects. Convolutional neural networks (CNN) are often used as a transfer learning model for image categorization. Insect Recognition Using Deep Transfer Learning Models Would Be Shown The IP102 and Leeds butterfly database, and insect classification were used for this study. In addition, a special database was established in Iraq called the Household Insects Database. Deep learning models Resnet50 and VGG19 were chosen as described in the study. The chosen models' robustness may be shown by performing tests that measure their accuracy and performance measures, including accuracy, recall, and F1 score, and concluding with a comparison of findings to other studies that have utilized the same IP102 dataset. Compared to other works in the field, the submitted work performed very well on every metric tested: precision, recall, and F1 score. The experimental results recommend using the Resnet50 model to detect insects because its accuracy reaches more than 95.11%. At the same time, the results of the vgg19 model came to 93.89%.