A0521
Title: Comparative study of the effectiveness of feature extraction methods for thermal image recognition
Authors: Yi-Ting Hwang - National Taipei University (Taiwan) [presenting]
Yu-Chen Yi - National Taipei University (Taiwan)
Bor-Shing Lin - National Taipei University (Taiwan)
Abstract: The development of artificial intelligence has expanded the applications of medical imaging. However, training artificial intelligence requires a large amount of medical imaging. Methods to enhance the accuracy of image interpretation are explored, even with limited equipment and image data. Thermal imaging is used as an example to explore three methods for extracting image features to identify nine poses. The first method is based on the features of the image itself, complemented by principal component analysis (PCA) and independent component analysis (ICA) to extract critical features. The second method employs spatial statistical models to characterize images using basis functions and considers the estimated coefficients as features. The third method combines the features of the first two methods. Six classifiers systematically compare feature extraction methods. Accuracy and F1 scores are used as the primary evaluation indicators of interpretative performance, and five-fold cross-validation is employed to ensure the stability of the classification results. The research results indicate that using ICA to extract feature variables demonstrated the best performance. Employing a spatial statistical model constructed with eight basis functions achieved optimal interpretation accuracy. Based on combining the two feature extraction methods, the accuracy and F1 scores obtained from the six classifiers were all above 0.85.
