http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
MC Dropout을 활용한 CNN 기반 악기 소리 분류의 성능 향상과 Out-of-Distribution 탐지
현준희(Junhee Hyeon),임채진(Chaejin Lim),한동일(Dongil Han) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
Convolutional neural networks (CNNs) are widely used in various fields, such as classification, object detection, segmentation, generation, natural language processing, and speech processing. Although CNNs exhibit strong performance on the trained data, they tend to fail on unseen data, leading to unexpected results. Therefore, it is essential to develop and research exception handling methods. In this study, we apply MC-dropout to the CNN model to handle exceptions and compare the performance with the model without MC-dropout. We evaluated the performance using a dataset consisting of instrument sounds, and different sounds. Image classification using CNNs is a wellknown method, but instrument sounds are represented as frequencies rather than images. Therefore, we convert sound into frequency to perform Image classification. We evaluated the ability to handle out-of-distribution data when MC-dropout is applied and examine its impact on the models performance. This study provides insights into improving the performance of instrument sound classification.