A Comparative Study of Preprocessing and Model Compression Techniques in Deep Learning for Forest Sound Classification

Abstract

Deep-learning models play a significant role in modern software solutions, with the capabilities of handling complex tasks, improving accuracy, automating processes, and adapting to diverse domains, eventually contributing to advancements in various industries. This study provides a comparative study on deep-learning techniques that can also be deployed on resource-constrained edge devices. As a novel contribution, we analyze the performance of seven Convolutional Neural Network models in the context of data augmentation, feature extraction, and model compression using acoustic data. The results show that the best performers can achieve an optimal trade-off between model accuracy and size when compressed with weight and filter pruning followed by 8-bit quantization. In adherence to the study workflow utilizing the forest sound dataset, MobileNet-v3-small and ACDNet achieved accuracies of 87.95% and 85.64%, respectively, while maintaining compact sizes of 243 KB and 484 KB, respectively. Henceforth, this study concludes that CNNs can be optimized and compressed to be deployed in resource-constrained edge devices for classifying forest environment sounds.

ESC-NAS: Environment Sound Classification Using Hardware-Aware Neural Architecture Search for the Edge

“…Notably, ESC-NAS models are well-suited for deployment on edge devices like the Raspberry Pi 3 B+ and other highly resource-constrained platforms such as microcontrollers, due to their minimal peak RAM usage and compact model size. Furthermore, by eliminating complex feature extraction procedures and spectrogram generation, the ESC-NAS models achieved reduced resource consumption and lower latencies compared to prior work employing models originally designed for image classification [ 38 ]. In summary, ESC-NAS models, specifically tailored for environmental sound classification on resource-constrained edge devices, present a promising solution, characterized by high performance and resource efficiency.…”

“…Therefore, the weights of the models trained during the ESC-NAS process were discarded and only their architectures were used without pretrained weights in the subsequent steps in the pipeline. Additionally, the predictive performance of the trained model was compared with seven selected models derived from the TinyML approach [ 38 ]. Furthermore, the trained model underwent deployment on the Raspberry Pi 3 B+ platform, where its RAM usage and inference time were tested.…”