Unveiling the Mystery: A Revolutionary Approach to Detecting Foreign Fibers in Long-Staple Cotton
Xinjiang long-staple cotton, renowned for its superior quality, is a cornerstone in the production of high-end textiles. However, the mechanical harvesting and processing methods often lead to the unintentional mixing of foreign fibers, such as plastic film, cotton boll hull, human hair, and polypropylene fibers. The current manual sorting process, while effective, is prone to human error and visual fatigue, impacting the accuracy and consistency of detection. Traditional identification methods, relying on color features and fluorescence reactions, fall short in distinguishing white, transparent, or color-similar foreign fibers, posing a significant challenge in the industry.
To address this issue, Associate Professor Ling Zhao and his team from the College of Mechanical and Automotive Engineering at Liaocheng University have developed an innovative solution. They introduced an intelligent identification method based on hyperspectral imaging and the PCA-AlexNet model, which has been published in the prestigious journal Frontiers of Agricultural Science and Engineering. This cutting-edge approach promises to revolutionize the sorting process, enhancing efficiency and automation.
The research team's approach is a game-changer. By integrating hyperspectral imaging technology with a deep learning model, they can simultaneously capture spatial and spectral information of objects. Each pixel contains reflectance data from multiple bands, forming a continuous spectral curve that can distinguish foreign fibers with similar colors. The team's innovative process begins with dimensionality reduction using principal component analysis (PCA) to select the optimal feature bands for each type of foreign fiber, reducing data redundancy and shortening model training time. They then fine-tuned the parameters of the classic AlexNet convolutional neural network, trained the model using the data from the selected feature bands, and finally determined the optimal model as PCA-AlexNet-23.
The experimental results are impressive. The PCA-AlexNet-23 model demonstrates exceptional performance in multi-class foreign fiber identification, achieving an overall accuracy (OA) of 97.2%, an average accuracy (AA) of 95.2%, and a Kappa coefficient of 93.1%. These metrics surpass traditional models like support vector machine (SVM), artificial neural network (ANN), and LDA-VGGNet. In practical sorting tests, the foreign fiber removal rate exceeds 85%, with the model excelling in identifying white, transparent, or color-similar foreign fibers, a challenge that traditional methods often struggle with.
The PCA technology plays a pivotal role in reducing the dimensionality of hyperspectral data while retaining the most critical feature information, avoiding interference from redundant data. The optimized AlexNet model, with its 2D convolutional structure, automatically extracts joint spectral and spatial features, enhancing classification accuracy. This approach is more efficient and accurate than 3D convolutional neural networks, which have numerous parameters and long training times.
Looking ahead, the research team aims to further expand the dataset of foreign fiber types, optimize data preprocessing technologies, and explore multi-source data fusion methods. Their goal is to continuously improve the performance of hyperspectral multi-target recognition algorithms, paving the way for the Xinjiang long-staple cotton industry to embrace efficient and fully automated mechanization. This breakthrough not only enhances the quality of cotton products but also contributes to the industry's sustainability and productivity.
