S. M. Hasan Sazzad Iqbal, Minhaz Ahmed Shikder, Md. Zohurul Islam, Md. Abu Johab, Most. Mira Khatun, Md. Niaz Imtiaz, Md. Shafiul Azam, Md. Abdur Rahim, Md Wadud Ahmed, Md. Toukir Ahmed
{"title":"Hyperspectral Imaging and Explainable Machine Learning for Broccoli Quality Assessment","authors":"S. M. Hasan Sazzad Iqbal, Minhaz Ahmed Shikder, Md. Zohurul Islam, Md. Abu Johab, Most. Mira Khatun, Md. Niaz Imtiaz, Md. Shafiul Azam, Md. Abdur Rahim, Md Wadud Ahmed, Md. Toukir Ahmed","doi":"10.1007/s12161-026-03340-0","DOIUrl":null,"url":null,"abstract":"<div><p>Broccoli, one of the most popular vegetables in the world, provides millions of people with a healthy source of nutrients. Its nutritional value, texture, and marketability are all influenced by its dry matter content (DMC). Therefore, accurate and timely mapping of DMC is crucial for broccoli quality assessment and decision-making in farming. In recent years, hyperspectral imaging (HSI) combined with machine learning techniques has become a powerful and reliable replacement for DMC prediction and crop characteristic evaluation. The goal of this research is to combine explainable machine learning models with visible-near-infrared HSI data (470–900 nm) to improve early-stage broccoli dry matter forecasting and enable image-driven agricultural investigation. A comparative analysis of predictive performance was conducted across three distinct machine learning architectures: random forest (RF), an ensemble tree-based method; convolutional neural networks (CNN), a deep learning approach; and partial least squares regression (PLSR), a linear regression model. The CNN model, with hyperparameters tuned via grid search, outperformed both the RF and PLSR models optimized using Bayesian optimization, achieving an <i>R</i><sup>2</sup> of 0.669, an RMSE of 0.90%, and an RPD of 1.73. Three distinct explainable artificial intelligence algorithms (Tree-SHAP, Kernel-SHAP, and DeepExplainer) were used to further analyze the predictions from the RF, PLSR, and CNN to identify the most significant wavelengths. The spatial distribution of the most crucial broccoli regions utilized to forecast dry matter was further visualized using the CNN model that was trained utilizing the significant wavelengths. The findings show how advanced agricultural analysis and precise DMC prediction may be achieved by combining HSI and explainable machine learning.</p><h3>Graphical Abstract</h3>\n<div><figure><div><div><picture><source><img></source></picture></div></div></figure></div></div>","PeriodicalId":561,"journal":{"name":"Food Analytical Methods","volume":"19 12","pages":""},"PeriodicalIF":3.0000,"publicationDate":"2026-11-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Food Analytical Methods","FirstCategoryId":"97","ListUrlMain":"https://link.springer.com/article/10.1007/s12161-026-03340-0","RegionNum":3,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"FOOD SCIENCE & TECHNOLOGY","Score":null,"Total":0}
引用次数: 0
Abstract
Broccoli, one of the most popular vegetables in the world, provides millions of people with a healthy source of nutrients. Its nutritional value, texture, and marketability are all influenced by its dry matter content (DMC). Therefore, accurate and timely mapping of DMC is crucial for broccoli quality assessment and decision-making in farming. In recent years, hyperspectral imaging (HSI) combined with machine learning techniques has become a powerful and reliable replacement for DMC prediction and crop characteristic evaluation. The goal of this research is to combine explainable machine learning models with visible-near-infrared HSI data (470–900 nm) to improve early-stage broccoli dry matter forecasting and enable image-driven agricultural investigation. A comparative analysis of predictive performance was conducted across three distinct machine learning architectures: random forest (RF), an ensemble tree-based method; convolutional neural networks (CNN), a deep learning approach; and partial least squares regression (PLSR), a linear regression model. The CNN model, with hyperparameters tuned via grid search, outperformed both the RF and PLSR models optimized using Bayesian optimization, achieving an R2 of 0.669, an RMSE of 0.90%, and an RPD of 1.73. Three distinct explainable artificial intelligence algorithms (Tree-SHAP, Kernel-SHAP, and DeepExplainer) were used to further analyze the predictions from the RF, PLSR, and CNN to identify the most significant wavelengths. The spatial distribution of the most crucial broccoli regions utilized to forecast dry matter was further visualized using the CNN model that was trained utilizing the significant wavelengths. The findings show how advanced agricultural analysis and precise DMC prediction may be achieved by combining HSI and explainable machine learning.
期刊介绍:
Food Analytical Methods publishes original articles, review articles, and notes on novel and/or state-of-the-art analytical methods or issues to be solved, as well as significant improvements or interesting applications to existing methods. These include analytical technology and methodology for food microbial contaminants, food chemistry and toxicology, food quality, food authenticity and food traceability. The journal covers fundamental and specific aspects of the development, optimization, and practical implementation in routine laboratories, and validation of food analytical methods for the monitoring of food safety and quality.