SAE-SOFTMAX-based Online Detection Method for Aquatic Anesthetics

Abstract

In view of the abuse of anesthetics in the transportation of aquatic products in vivo, this paper proposes a real-time online monitoring method for aquatic anesthetics based on the SAE-SOFTMAX model. The model herein normalizes the collected physical and chemical parameters of aquatic anesthetics, extracts data features through stacked sparse autoencoder (SAE), uses feature data and category tags to supervise and train the SOFTMAX classifier, and finally builds a prediction model to achieve classified prediction of aquatic anesthetics varieties. Experimental research results prove that the method proposed herein has better prediction accuracy and generalization performance. The use of feature data to train the SOFTMAX classifier can improve the prediction accuracy, and the SAE-SOFTMAX deep network can be finely-tuned to make the prediction accuracy up to 98.08%.