Precision Agriculture for Pest Management on Enhanced Acoustic Signal Using Improved Mel-Frequency Cepstrum Coefficient and Deep Learning

D. Poornima and Dr. G. Arulselvi

In the recent past, the agriculture sector has become smart by utilizing of modern technologies. In traditional agriculture, toxic pesticides are used to control pests. These pesticides also kill the beneficial insects in pollination for enhancement of productivity. Nowadays, many technologies have been emerged to monitor and control the pest activities, but all these techniques may time consuming, causeā€Ÿs damages to crops etc. In this paper, a novel algorithm Precision Agriculture for Pest Identification and Classification (PA-PIC) is proposed to identify and classify harmful pests and beneficial insects in the greenhouse farming without damaging the crops. The fundamental frequency of different types of pests and insects are estimated for PA-PIC using acoustic sensor and recording system. The noise occurred along with acoustic signal should be properly filtered. The adaptive audio signal enhancement is applied to restore and enhance the acoustic signal to improve quality and adaptive filters remove entire noise harmonics from noisy acoustic signal. The Harmonic Noise Model (HNM) Wiener filter, HNM HMM Wiener filter and HNM Infinite filter are comparatively used to remove the noise harmonics from the source acoustic signal based on its performance. The acoustic signal segmentation and feature extraction are used for classification using Deep Learning (DL) technique. The entire activities of PA-PIC are given to IoT to monitor by the farmer. Based on the performance evaluation of spectrogram and objective measures, the HNM HMM Wiener filtering algorithm is achieved significant improvement over conventional methods at reduction in noise levels. The features of acoustic signals are calculated using feature learning for further classification using Improved Mel- Frequency Cepstrum Coefficient (IMFCC). The Deep Neural Network (DNN) is applied to classify whether then given signal is pest or insect. The experimental results are proven that the proposed methodology is achieved high efficiency and accuracy by graphical representation and tabulation.

Volume 12 | 03-Special Issue

Pages: 50-65

DOI: 10.5373/JARDCS/V12SP3/20201238