An efficient feature pyramid network with adaptive LSTM for pest detection and classification in IoT

Abstract

Abstract Crop pests are a major cause of economic loss and environmental damage globally. Timely detection of pests is crucial for protecting crops and maintaining the global food supply. However, existing diagnostic methods are especially manual, demanding significant time and expert knowledge. Incorrect pest identification can result in the misuse of pesticides, affecting both crop yields and the surrounding ecosystem. Therefore, there is a need for an automated solution that offers more precise pest identification and classification. So, in this research work, a new Internet of Things (IoT)-based pest detection and classification technique is implemented. In the initial phase, essential images are collected from a standard database that includes the IoT sensor-based pest images. Next, the IoT sensor-based images are offered as the input to the Joint pest detection and classification phase. In this phase, a new framework named Feature Pyramid Network with Multi-Attention Fusion Vision Transformer-based Adaptive Long Short Term Memory (FPN-MAFViT-ALSTM) is employed to execute the pest detection and classification procedure. Moreover, parameters in FPN-MAFViT-ALSTM are tuned using Enhanced and Intelligent Gooseneck Barnacle Optimization with Randomized Exploration (EIGBO-RE), which helps in improving pest detection and classification. At last, pest detection and classified outcomes are obtained from FPN-MAFViT-ALSTM, and then various experiments are carried out to verify its efficiency under varying conditions.