BDL-Net: A blended deep learning approach for pest detection in agriculture using IoT-enabled sound analysis

Agriculture around the world has developed to a stage where it’s a profitable distribution of high-tech farming skills. This is a turning point in agricultural history. This change was made possible due to the popularity of information and its emerging technologies. Yet the pests of crops are the main destroyers of arable land, with deleterious consequences for the economy, the environment, and society. This demonstrates the importance of automated technologies in identifying these pests before they cause substantial damage. Recently, ML-based research has primarily been related to agriculture. The study’s main objective is to use advanced pest sound analytics and IoT to offer a practical means of monitoring pests within large agriculture areas. The proposed system comprised several adequate analytics-based audio pre-processing methods. These were the PLP, spectral centroid algorithms, FFT, DFT, Hann window, Wiener filter, and the Hamming window. Training, using 3,000 pest sounds of 30 pests tested for their features and statistical information, was implemented with POS in the Pyramid Attention Graph-Based Multi-Module DCNN (BDL-Net). The CSPNet, CTA, MPF, GPE, and RegNet modules are just several examples of the many critical auxiliary modules integrated into the BDL-Net framework. At an accuracy of 99.86%, sensitivity of 99.94%, specificity of 99.61%, recall of 99.94%, precision of 99.87%, F1 score of 99.91%, the results demonstrated that the proposed BDL-Net model predicted the pest with high accuracy and outperformed the existing DenseNet, DCNN, ResNet-50, YOLONDD, YOLOv8, DMAUS-Net, and Pest-PVT, Modified CBAM, DB-CAFNet, and DNN models. This study is essential as it can detect pests in vast agricultural regions at an early stage. Crop yields will increase, helping to drive both the country’s and the world’s economic growth and that of farmers.