Application of SVM Based on Optimization of Newton Raphson's Algorithm in Non-Invasive Blood Glucose Detection

Traditional invasive blood glucose monitoring methods carry risks such as wound infections and patient discomfort. To address these issues, we propose a non-invasive method based on facial infrared thermography, aiming to enhance patient comfort and improve the accuracy and convenience of blood glucose detection. To address the data imbalance problem, a wavelet-based sample pairing fusion technique was used to enhance the thermal imaging dataset. Features extracted by the MobileNetV3 network were input into an SVM model for training, and the Newton–Raphson optimization algorithm was applied to optimize the SVM parameters to improve performance. Compared with the standalone MobileNetV3 model, the MobileNetV3-NRBO-SVM regression network exhibits better performance in terms of maximum error and root mean square error (RMSE). The predicted blood glucose values of our proposed model are all within region A of the Clark error grid with a maximum deviation of less than 10%. These results indicate that the non-invasive blood glucose detection technique based on infrared thermography and the MobileNetV3-NRBO-SVM model proposed in this study achieves clinically acceptable accuracy.