Study on the State Evaluation of Novel Nanocomposite Waterborne Polyurethane Material Based on Image Analysis and Improved U-Net Neural Network
Ying Wang · 2024
To further improve the precision and reliability of material status assessment, the new nanocomposite waterborne polyurethane is taken as the research object, and a material state assessment method based on improved U-Net neural network is proposed. Firstly, images of research materials are collected and preprocessed. Then, U-Net neural network is used as the basic image classification recognition method, and attention mechanism and residual network are adopted to improve it, so as to improve its feature extraction ability. Finally, the improved U-Net neural network is utilized for precise evaluation of the state of new materials. Experimental results show that compared with FCN network, U-Net neural network after improvement can converge faster to smaller loss values. When performing state recognition on materials, the improved U-Net neural network has higher precision, with accuracy, precision, recall and IOU reaching 0.980, 0.981, 0.971 and 0.969 respectively. Compared with PCA-GA-BP method and CNNSVM method, the evaluation precision and efficiency of the proposed method are better. This shows that this method can achieve the accurate state evaluation of the new nanocomposite waterborne polyurethane, and can be applied to the actual material state assessment scenario, which is highly feasible.