IoT-based Pest Detection Using YoloV5 with Trans-Unet and Classification via Adaptive Mobilenet in Agricultural Fields

Abstract Applications of Internet of Things (IoT) in agriculture have been incredibly succeed in raising agricultural output. It faces challenges like declining crop yields and climatic hazards with frequent and severe insect and illness outbreaks. Predicting insect outbreaks with precision can be essential to raise agricultural productivity. The following factors are crucial for predicting pests: humidity, rainfall, temperature, velocity of the wind, and length of sunlight. A model using deep learning that generates binary choices regarding the existence of insect populations depending on environmental circumstances is fed sensed variables from the environment. IoT-based identifying pests and categorization systems have a lot of potential, but it has drawbacks, including the requirement for widespread deployment, privacy concerns, and reliable operation. It requires a combination of creative technology solutions to overcome these shortcomings. Hence, an improved IoT-based pest detection model is developed in this proposed work. The image needed for this proposed model is collected from the standard database and fed to the detection. In this phase, the pest is detected using You Only Look Once V5 with TransUnet (YoloV5-TUnet). Early detection of pests helps farmers to take appropriate and targeted actions to control pests before it causes significant damage to crops. The detected pest is further applied to the classification, which is done by Adaptive MobileNet (A-MNet). The parameters of A-MNet are tuned by Improved Frilled Lizard Optimization (IFLO) to enhance the classification performance. Classification of pests helps in the early identification of invasive species that cause threats to crops and ecosystems. The success rate of the system is validated and measured across divergent metrics. The experimental outcome of the proposed model shows superior performance than other existing models.