A High Accuracy Deep Learning and Iot Network for Pest Detection Using Analysis of The Pest Sound in Large Agriculture Area

The significance of agriculture in ensuring human survival in the contemporary day necessitates enhancing crop yields. Pest infestations have the potential to cause harm or hinder the growth of crops, hence exerting a negative impact on agricultural productivity. Therefore, it is imperative to initiate agricultural pest identification and management measures promptly. Pest surveillance poses significant hazards, requires extensive physical effort, and is labor-intensive. The combination of emerging AI and IoT technologies has the reliability of pest identification in agriculture. These advancements enable autonomous control of maintenance chores, leading to improved outcomes in pest management. The present work introduces an innovative technique for the real-time identification of pests by leveraging the IoT and DL algorithms. The pest identification technique DConvNet was trained using pest audio sound, preprocessed by Planck-taper filter, DPSS approach, non-overlap-add method, and features extracted by LPCC and GMM methods. This study illuminates the aspirations of the IoT and AI in pest monitoring in agricultural settings, with the benefit of minimizing the need for human involvement to maintain constant vigilance. The DConvNet system demonstrated high accuracy in automating pest identification, as evidenced by extensive agricultural field experiences. The proposed model achieved superior performance compared to KNN, ANN, VGG-16, and EfficientNet B3 in the task of pest detection, achieving an accuracy of 99.72%, recall (99.78%), and F1 score (99.86%).