IoT-based pest detection and classification using deep features with enhanced deep learning strategies

Agriculture is an essential source of sustenance. Here, pest detection is very helpful in increasing food quality, and also it helps to increase the country's economy. Moreover, the rapid infestation of pests and insects may face a serious problem in agricultural yield. In existing works, the images are collected from the IoT sensors for pest detection and classification, but it is not satisfactory to acquire higher results regarding accuracy. To solve this issue, a novel pest detection and classification model is suggested. The major intention of the developed method is to detect and identify pests over crops using an object detection model and enhanced classifier. Initially, the IoT platform is included, where the required data is garnered through agriculture-based IoT sensors. The input images are garnered and fed into the subsequent object detection process. Pest detection is obtained through an optimized Yolov3 model. Here, the hidden neurons are optimized by the Adaptive Energy -based Harris Hawks Optimization (AE-HHO) algorithm. The detected pest images are given to the feature extraction process. The deep feature extraction is utilized with the help of Residual Network50 (ResNet50) and Visual Geometry Group16 (VGG16) models. Then, the classification is performed by Weight Optimized Deep Neural Network (WO-DNN) model. Consequently, the WO-DNN of the weight factor is optimized using the AE-HHO algorithm. Finally, the classified outcome is predicted by the AE-HHO algorithm. Hence, the performance is validated with several measures and compared with existing detection methods. Throughout the analysis, the accuracy and F1-score of the designed method attained 96% and 84%. Thus, the results explore that the suggested method proves better efficiency in pest detection and classification.