A Cloud-hosted web app utilizing a hybrid approach for insect and mold classification

A Cloud-hosted web app utilizing a hybrid approach for insect and mold classification

연구 분야: Software Development

학회: Multimedia Tools and Applications

Insects and molds pose a serious threat to stored products, particularly grains. Under favourable conditions, these infestations can proliferate rapidly, and late discovery can lead to contamination, affecting both the quality and quantity of the products. To address this issue, several automated systems have been proposed, but they often face challenges such as insufficient data, sensitivity to noise and harsh environments, and feasibility for real-world application. Therefore, there is significant scope for improvement in making these applications more precise, robust, and accessible to users. This study proposes a web application implementing a hybrid approach that combines audio and image classification models to provide accurate and reliable real-time classification predictions via a cloud-hosted web application. A CNN was constructed for the audio classification of insects, and a YOLOv8 classification model was implemented for insect and mold images. The predictions from these models were combined using the concept of late fusion, and the performance of this combined approach was compared with the individual modalities. Finally, the proposed approach was integrated into a Flask web application, deployed using Docker on an AWS EC2 instance, to provide users with an accessible platform for early-stage detection and classification.