AI-Based bread quality assessment using image processing techniques and the developed BQe-CNN

Authors: SERFA JUAN, RONNIE - Orise Fellow; Kaufman, Rhett; Bean, Scott; Chen, Yuanhong; Sutton, Theresa; ABACAN, SHEILA - Kansas State University; Gerken, Alison; Tilley, Michael

Submitted to: International Journal of Food Science and Technology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/18/2026
Publication Date: 1/23/2026
Citation: Serfa Juan, R., Kaufman, R.C., Bean, S.R., Chen, Y., Sutton, T.L., Abacan, S.F., Gerken, A.R., Tilley, M. 2026. AI-Based bread quality assessment using image processing techniques and the developed BQe-CNN. International Journal of Food Science and Technology. https://doi.org/10.1093/ijfood/vvag010.
DOI: https://doi.org/10.1093/ijfood/vvag010

Interpretive Summary: Bread quality plays a vital role in consumer satisfaction, yet traditional quality assessment methods often rely on costly equipment (such as the C-Cell Baking Quality Analyser) or subjective scoring by trained experts. This leads to lag times in determining whether a wheat variety is suitable for bread baking, which can result in delays in advancement of wheat breeding lines. To address this challenge, this study introduces the Bread Quality Enhanced Convolutional Neural Network (BQe-CNN)—a specialized artificial intelligence model designed to classify key bread quality attributes such as porosity, texture, and color with high precision based on image analysis. These models were trained on three years of color quality and crumb scores from experts in the baking industry from a range of wheat varieties bread quality scoring data. The model achieved a classification accuracy of 96% on testing data, outperforming other machine learning approaches including Decision Tree, Random Forest, and Support Vector Machines. Beyond classification, BQe-CNN offers valuable insights into the relationship between flour characteristics, baking parameters, and final product quality, allowing us to predict bread quality based on other variables, beyond image data. This allows wheat breeders and food scientists to better assess bread making potential of new wheat varieties earlier on in the production chain, bypassing the need for expensive machines like the C-Cell and reducing the reliance on human subjectivity while enhancing throughput.

Technical Abstract: Ensuring consistent bread quality is vital for maintaining industry standards, reducing waste, and ensuring consumer satisfaction. Traditional methods of bread quality analysis rely on expertly trained manual inspection are often subjective, time-consuming, and prone to inconsistencies, while modern analysis techniques using proprietary machines, though available, tend to be prohibitively expensive. This study introduces an AI-driven approach that leverages advanced image processing techniques to automate and enhance the accuracy of bread quality assessments. By extracting key features such as porosity, texture, and air cell structure, and using expert-scored color and crumb score metrics, the proposed Bread Quality Enhanced Convolutional Neural Network (BQe-CNN) offers a precise analysis of bread parameters. The model achieved classification accuracies of 92% for bread colors and 88% for quality levels. These results suggest that these methods could significantly streamline manual scoring methods, reducing the need for expertly trained, subjective evaluation or high-priced machines. By leveraging enhanced layers like residual connections and attention mechanisms, the model efficiently captured fine details in bread images, making it highly effective at detecting subtle variations in texture and air cell distribution. While the model demonstrates high performance in quantitative analysis, it is important to note that artisan scoring—characterized by detailed aesthetic evaluations integral to traditional bread-making—remains a challenging domain for automation. This limitation presents an opportunity to further enhance the model's capabilities by integrating advanced algorithms or hybrid approaches, bridging the gap between precise computational analysis and the specific requirements of artisan scoring. Nevertheless, the BQe-CNN's ability to provide real-time, automated quality control is a dependable and transformative tool, optimizing production, reducing waste, and complementing human expertise in a cost-effective manner. This novel approach, rooted in visual analysis of product characteristics, represents a significant leap forward in achieving consistency and scalability in bread quality control for the baking industry.