Browsing by Author "Arias Rivas, Betzabeth Abigail"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Computer vision algorithms for characterizing and classifying agro-exportable grains through size analysis and clustering techniques in Peru's agricultural sector
    (Universidad Nacional de Cañete, 2026-03-06) Arias Rivas, Betzabeth Abigail; Machaca Mamani, Julio Cesar; Cahuana Lipa, Rocio; Manrique Nugent, Manuel Alberto Luis
    The implementation of machine learning and deep learning algorithms has revolutionized quality control processes in the agroindustry, providing precise and efficient solutions for grain classification and evaluation. The use of computer vision, combined with advanced algorithms, enables defect detection, measurement of physical characteristics, and optimization of batch standardization for both domestic and international markets. Objective: To review and analyze the application of ML algorithms in grain classification through digital image processing, comparing their accuracy, efficiency, and processing time to identify the most suitable strategy for productive environments. Materials and Methods: A systematic review was conducted in Scopus, IEEE Xplore, ScienceDirect, and MDPI, considering publications from 2017 to 2024. Inclusion criteria were: use of RGB or hyperspectral images, application of CNN, SVM, K-means, or Random Forest algorithms, and reporting of quantitative metrics. Review articles, studies not applied to grains, or those without experimental validation were excluded. Results and Conclusions: CNN achieved the highest accuracy (≈97%), making it ideal for detailed classification, although it requires high computational resources and longer training times. SVM and Random Forest demonstrated a balance between accuracy (≈91–92%) and efficiency, making them suitable for medium-sized datasets. K-means stood out for its speed and simplicity, although with lower accuracy (≈88%) as it is an unsupervised method. Emerging trends point to the integration of hyperspectral vision, transfer learning, and hybrid approaches to optimize the balance between accuracy, speed, and operational feasibility, thereby enhancing the competitiveness of the agroindustry in global markets.