JoWUA

Abstract

Convolutional neural networks (CNNs) have become the most widely used algorithms for image recognition and analysis, as they can automatically learn data features. Commonly, CNNs are of interest in the sensor data analysis area, with specific attention to human physiological data measured by sensors on wearable devices. A specific physiological signal to which the information might be ink-related is the perspiration rate, which includes health or stress-related information. Response to: An optimized CNN model for perspiration rate sensing in wearable electronics. The objective here is to optimize the CNN model with minimal computational effort and yet deliver higher accuracies with efficiency. We address specific issues in physiological data, such as noise and variance, and develop a dedicated CNN architecture by incorporating our modifications to the algorithm for enhanced robustness. Namely, the model comprises convolutional and pooling layers to extract features from the physiological signals, as well as fully connected layers for classification. To improve the performance of our model, we proposed several optimization approaches. The batch normalization layer is implemented to accelerate convergence and suppress overfitting. Transfer learning is a deep learning method that utilizes pre-trained CNN models as an initial point and then fine-tunes the model according to our dataset. We also employ a data augmentation method that enhances the model's ability to generalize to unseen data. These strategies aim to achieve high precision and efficiency in terms of perspiration rate sensing. We evaluate our DNS-CNN + BO model using a dataset collected from wearable sensors. Experiments show that our method achieves state-of-the-art performance compared to methods in terms of accuracy and computing cost. Additionally, we compared our models with classical machine learning methods, which shows that CNNs are more effective in capturing features and processing physiological signals.