Wavelet & Transfer Learning Based Prosthetics Gestures Control

The combination of Discrete Wavelet Transform denoising and transfer learning provides an effective approach for the categorization of electromyography data in prosthetic control. Electromyography data, frequently influenced by noise and complexity, are enhanced using discrete wavelet transform denoising, which separates essential characteristics while attenuating interference. The denoised signals are transformed into images via Continuous Wavelet Transform and subsequently input into a fine-tuned pre-trained AlexNet deep learning model for feature extraction. The characteristics are utilized to train a shallow classifier, forming a hybrid model that integrates the advantages of wavelet preprocessing with transfer learning. The model attains a test accuracy of 94.4% and a training accuracy of 100%, surpassing conventional approaches in both precision and efficiency. This scalable and adaptable system offers potential for real-time prosthetic control, improving the reliability and functionality of assistive devices.