Development of a Novel IoT–UAV‐Based Early Detection and Identification of Pests Using Optimized Deep Learning Network

ABSTRACT The daily needs of human society depend substantially on food crops, which are constantly threatened by pests. The majority of people involved in agriculture focus on pest management for both financial success and social survival. The proper control of pests is important for human society, not only for farmers. However, a number of factors like emergencies, severe downpours, floods, earthquakes, pest damage, etc. makes it extremely difficult for farmers to safeguard crops after as well as during production. So, identifying the pests in the crop is crucial to take proactive measures for its protection. Therefore, in this work, the pest detection and classification model is developed based on the Unmanned Aerial Vehicle (UAV)‐assisted IoT that enables early‐stage pest detection with higher accuracy. The UAV is used to frequently monitor the crop field and is especially helpful for covering large agricultural areas. The IoT‐based image collection is initially performed, and these images are applied to the pest detection model, where the Adaptive and Attention‐based You Only Look Once (Yolo) V9 with Novel Loss (AA‐YoloV9NL) is implemented for pest detection. YoloV9 is highly suitable for object detection tasks, particularly for identifying small‐sized pests, and hence, this method is chosen in this study. To improve the detection efficacy, the Lemur‐Inspired Fossa Optimization Algorithm (LI‐FOA) is applied to optimize the hyperparameters present in the AA‐YoloV9NL. The detected pests are processed by the Pyramidal Dilated Residual Densenet (PRDNet) model for classifying the types of pests. Based on the identified pest types, appropriate control measures are applied during crop production, thereby avoiding unnecessary pesticide use in agricultural lands. The proposed model's effectiveness is evaluated by validating the results. When considering the linear activation function, the accuracy of the proposed framework is 96.05%, which is 10.9%, 13.01%, 11.4%, and 8.2% higher than the existing approaches like YOLOv5, Tiny Yolov4, YO‐CNN, YoloV9, respectively. The findings confirmed that the introduced framework can effectively minimize the labor cost and the use of hazardous chemicals to boost the effectiveness of the agricultural practices.