Deep learning and computer vision for image-based high-throughput phenotyping of canning quality traits in dry beans

Authors: SINGH, LOVEPREET - Michigan State University; MADHAV LELLA, OM SAI KRISHNA - Michigan State University; WRIGHT, EVAN - Michigan State University; Cichy, Karen; HOYOS-VILLEGAS, VALERIO - Michigan State University

Submitted to: The Plant Phenome Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/14/2026
Publication Date: 4/23/2026
Citation: Singh, L., Madhav Lella, O., Wright, E.M., Cichy, K.A., Hoyos-Villegas, V. 2026. Deep learning and computer vision for image-based high-throughput phenotyping of canning quality traits in dry beans. The Plant Phenome Journal. 9(1). Article e70080. https://doi.org/10.1002/ppj2.70080.
DOI: https://doi.org/10.1002/ppj2.70080

Interpretive Summary: The canning quality of dry beans is a critical factor determining the value and marketability of the crop in US domestic markets. Traits such as visual appearance, color, and texture are the primary criterion to assess the canning quality of a dry bean cultivar. The complex interaction between genetics, processing methods and environmental conditions makes canning quality assessment an important component of modern dry bean breeding programs. Current phenotyping methodologies rely on two major approaches for evaluating canning color in dry beans: visual rating and instrumental colorimetry. Both are time consuming and prone to errors. To overcome the canning quality phenotyping bottleneck, this study aimed to develop an open-source, high-throughput pipeline tailored for canning quality trait evaluation in dry bean breeding programs. Using digital images of canned bean samples from Michigan State University's dry bean breeding program, this workflow integrates standardized image acquisition of canned bean samples with statistical, computer vision, and deep learning tools to deliver a streamlined pipeline that processes the images for bean detection, segmentation, pixel level color extraction and delivers an objective score metric. This pipeline was developed to run on local computers as well as high performance-computing clusters, making it suitable for processing large batches of images of canned bean samples in a practical large breeding program. The major advantages and contribution of this workflow are a novel objective metric for dry bean canning color evaluation which can significantly decrease cost, reduce subjectivity and bias in the analysis and an objective color metric that statistically outperformed the other canning color evaluation methods such as visual rating and instrument colorimetry. The workflow for canning color evaluation can enhance the selection accuracy of breeding materials and consequently, increase the rate of genetic gain and can be expanded to the other high-throughput objects instance segmentation such as texture, seed size, seed shape and canning appearance.

Technical Abstract: Canning color retention is a key quality trait in dry bean breeding, influencing consumer acceptance and commercial value. Public breeding programs maintain canning quality as a selection trait of importance, but existing color evaluation methods such as visual rating are subjective, while instrument colorimetry is costly, provides limited throughput, and often struggles to accurately detect differences between genotypes. To address these challenges, we developed a high-throughput phenotyping pipeline that integrates computer vision and deep learning to improve canning color quality assessment in dry beans. This pipeline combines the YOLOv8n object detection model with the segment anything model (SAM) for precise bean segmentation. 525 black dry bean genotypes from Michigan State University preliminary and advanced yield trials over multiple years were evaluated using this pipeline. The final model was trained with over 1200 images of canned dry bean samples. The YOLOv8n model achieved near-perfect detection performance, with precision reaching 0.99982 and recall reaching 1 after 44 epochs of training. Comparative analysis showed that the image-derived D-Score consistently outperformed visual ratings and colorimetry methods, with lower prediction error in regression models. The D-Score metric offered high resolution in color assessment, enabling the distinction of subtle, genotype-level differences. Additionally, the D-Score provides a ranking criterion that enables breeders to make informed decisions and select superior genotypes. This pipeline was deployed to Michigan State University High Performance Computing Center (HPCC) and can process 200 high-resolution images in under five minutes, making it practical for large-scale breeding applications while eliminating subjective bias and reducin