Research On Deep Learning Method For Mixed Dye Concentration Detection

Printing and dyeing play vital roles in the quality control process of textile processing.Various factors impact the success rate of printing and dyeing,including precise dye concentration,pH values,temperatures,drying temperature,and the quality of dyeing auxiliaries.Among them,accurate preparation of dye concentration is a core process.To address the limitations of traditional dye concentration detection methods,such as low accuracy and efficiency,this study utilizes industrial camera devices to capture and extract RGB values of the dyeing solution under specific lighting conditions.By integrating data processing techniques,deep learning algorithms,and iterative methods,this thesis presents different concentration detection methods for three-component mixed dye solutions in different textile application scenarios.The main contents are as follows:(1)A novel approach is presented in this study,utilizing a CNN-BiLSTM architecture to address the challenges in detecting the concentration of three-component mixed dye solutions.This method is precious in scenarios requiring high accuracy,such as color innovation research and batch production to ensure consistent quality.By leveraging the temporal sequence characteristics of RGB values obtained under varying exposure times,the proposed method employs a BiLSTM neural network algorithm along with a CNN(Convolutional Neural Networks)for feature extraction.This enables the establishment of a concentration prediction model based on CNN-BiLSTM.To enhance the stability of the training process,two iterative methods are incorporated to establish linkages among the three dye concentration prediction models.The inclusion of these linkages results in average prediction accuracies of 97.89%and 98.05%,respectively.(2)A CNN-GRU-based segmented concentration detection method is proposed.This method addresses scenarios where accuracy and efficiency are both important,such as daily monitoring of industrial dyeing tanks,educational experiments in the textile printing and dyeing industry,and QC quality control.The method utilizes a set of small samples with high similarity and employs a lightweight GRU(Gated Recurrent Unit)as the algorithm foundation for establishing a dye concentration prediction model based on CNN-GRU.Experimental results demonstrate improved overall efficiency compared to the current highest accuracy model,while maintaining higher detection accuracy than traditional machine learning methods,striking a balance between accuracy and efficiency.(3)To facilitate the use of the prediction models,this thesis designs and implements a dye solution detection system using the Vue+Django technology framework.The proposed prediction algorithms are integrated into the dye solution detection system,providing users and enterprises with simple and easy-to-use dye concentration detection methods.