Abstract:

Objective Color fastness grades evaluate how well textiles maintain their colors under different conditions. Traditional visual ratings of textile color fastness can be inconsistent and often vary from the measurements obtained through spectrophotometers, which can lead to discrepancies with human visual perception. To address these challenges, a new image-based color measurement method has been proposed for assessing textile color fastness more accurately. This study aims to create a method for grading textile color fastness by using image-based color measurement technology and BP neural network. It also seeks to enhance the performance of the traditional color difference conversion method for better application in textile color fastness grading. The importance of this research lies in introducing a new technique for digital image-based color measurement, which involves using spectral reconstruction to calibrate the digital camera and colorimetry theory to calculate the color of textile samples. Additionally, the BP neural network is utilized to model the color information difference as input and expert grading results as output.

Method Firstly, three groups of color fastness tested samples were made according to the recommended standards, where a total of 341 sets of staining samples and 125 sets of discoloration samples were produced in the experiment. Among them, the staining samples included 155 sets of rubbing samples and 186 sets of washing samples, and 125 sets of discoloration samples were obtained through exposure experiments. After that, using the self-developed imaging-based color measurement system, the spectral reflectance of samples is obtained by spectral reconstruction technology, and the color information of the sample is then calculated by colorimetry theory under CIED65 illuminant and CIE1964 standard observer color matching functions. Then, through psychophysical experiments, color fastness grading experts were invited to grade all color fastness samples, and the expert grading results of all test samples were obtained. With the color fastness grading results from experts on sample pairs as references, the BP neural network was used to construct a relationship between the color difference of sample pairs and the color fastness grades of experts. Finally, for any new given test sample pair, the constructed BP network was used to predict color fastness grade and compare it with the expert grading results. At the same time, the traditional color difference conversion method for color fastness grading was also optimized by introducing correction strategies and models.

Results The experimental results showed that the prediction root mean square error(RMSE) of the BP neural network method for staining and discoloration samples were 0.38 and 0.41, respectively, which are significantly better than the curve fitting method and the color difference conversion method before correction. The prediction error of the color fastness of the staining samples was found slightly higher than that of the corrected color difference conversion method. In addition, the prediction RMSE of the BP neural network method for friction and soaping samples were 0.30 and 0.27, respectively, and the prediction error was lower than that of the corrected color difference conversion method. Overall, the prediction performance of the corrected color difference conversion method was significantly better than that before correction. The prediction effect of the BP neural network was consistent in general with that of the corrected color difference conversion method, and was significantly better than the curve fitting method.

Conclusion This study has led to the establishment of a color fastness prediction method based on the digital image-based color measurement technology and BP neural network. In addition, an optimization correction strategy is proposed to improve the performance of the color difference conversion method. The effectiveness of the proposed BP neural network method and the optimized color difference conversion method are proved through actual experiments. Compared with the spectrophotometer-based method, the new method is more practical and flexible. The research findings provide useful information for promoting the digital development of textile enterprises and have application value.

Key words: textile, color fastness prediction model, imaging-based color measurement, spectral reconstruction, colorimetry theory, color fastness grading, BP neural network