Enhancing nutrient content estimation in lettuces using hyperspectral image data and artificial neural networks with feature selection methods

Authors: ESHKABILOV, SULAYMON - North Dakota State University; Simko, Ivan; NEHA, FARHIN - North Dakota State University; PRANTO, MAHMUD ALAM - North Dakota State University; SIMSEK, HALIS - Purdue University

Submitted to: Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 4/11/2025
Publication Date: 7/26/2025
Citation: Eshkabilov, S., Simko, I., Neha, F., Pranto, M.A., Simsek, H. 2025. Enhancing nutrient content estimation in lettuces using hyperspectral image data and artificial neural networks with feature selection methods. In: Kahraman, C., Cebi, S., Oztaysi, B., Onar, S.C., Tolga, C., Sari, I.U., Otay, I., editors. Artificial Intelligence in Human-Centric, Resilient and Sustainable Industries, Proceedings of the INFUS 2025 Conference, Volume 2. International Conference on Intelligent and Fuzzy Systems, July 29-31, 2025, Istanbul, Turkey. p. 234–242. https://doi.org/10.1007/978-3-031-97992-7_27.
DOI: https://doi.org/10.1007/978-3-031-97992-7_27

Interpretive Summary: Lettuce is a widely consumed leafy vegetable in the U.S., playing a significant role in healthy diets. While traditionally grown in open fields, there's a growing shift towards controlled environment agriculture. This study explored using hyperspectral imaging combined with artificial neural networks (ANNs) to learn if this combination could accurately and quickly determine the nutrient composition and overall quality of lettuce grown in controlled environments, potentially reducing the need for manual laboratory work. The ANN system was able to accurately classify different aspects of the lettuce and could estimate its nutrient content. The comparison of different information selection methods from the hyperspectral data also provided valuable insights. This research suggests that using advanced imaging and artificial intelligence could significantly improve how lettuce composition is checked. This technology offers the potential for faster, more efficient, and less labor-intensive quality control in the lettuce industry.

Technical Abstract: Lettuce is a crucial component of a healthy, well-rounded diet. The annual consumption of leafy vegetables in the U.S. is approximately 6 kg per capita. While most lettuce crops are traditionally grown in open fields, there has been a recent increase in production within controlled environment systems. U.S. lettuce growers are currently facing several challenges, including labor shortages and rising costs, water scarcity, high fertilizer costs, and concerns about food safety. Lettuce production and quality control processes are labor-intensive. Traditionally, the nutrient composition of lettuce has been determined through laboratory tests conducted by trained technicians. However, advancements in technology, such as computer vision and digital imaging, offer the possibility of obtaining real-time data while reducing labor costs. This study utilized hyperspectral image data and artificial neural networks (ANN) to estimate the nutrient composition and quality of lettuce grown in a controlled environment. One challenge in the application of ANN and other machine learning algorithms is selecting the most appropriate features to prevent overfitting in predictive model development. To address this, various feature selection and data size reduction methods were explored. The results of the study indicate that the ANN model accurately classified lettuce contents at a rate of 100% and estimated nutrient composition within the range of 0.85 to 0.99. Different feature selection approaches from the hyperspectral image data were compared, including first-order derivatives, principal component analysis, partial least squares regression, multivariate regression, and variable importance projection.