Design of an Intelligent Ecological Environment Monitoring System Based on Sound Analysis

This study proposes an intelligent ecological environment monitoring system based on Long Short-Term Memory (LSTM) networks, aiming to achieve automatic identification of sounds from 15 commonly encountered animal species in Taiwan's urban and streamside areas. To accomplish this goal, the system utilizes Mel Frequency Cepstral Coefficients (MFCC) as the method for extracting audio features, feeding preprocessed audio data into the LSTM model for training. The model structure comprises two layers of LSTM units, incorporating dropout layers and early stopping mechanisms to prevent overfitting. Experimental results indicate that the model achieves a classification accuracy of 91.01 % on the test set with a loss value of 0.3309, demonstrating excellent generalization ability. This system can be applied to ecological monitoring in and around Vanung University, utilizing sound recognition technology to monitor the ecological environment in real-time, assessing biodiversity changes, particularly in evaluating ecological restoration following local environmental remediation. Moreover, the system holds potential value for teaching and scientific research, providing concrete practical applications for ecological monitoring and data analysis within the university.