A Machine Learning Framework for Wheat Seed Variety Classification Using GLCM-Based Texture Analysis and Feature Optimization

In Pakistan, wheat is one of the important crops, and the production of wheat is highly reliant upon the use of pure and certified seed varieties. Hence, wheat seeds should be properly identified to ensure the quality of the crop and better production. Farmers typically rely on a visual assessment of the varieties of seeds in practice. This technique is incorrect and prone to inaccuracies. Many seeds of wheat are very similar in shape and color. This wrong seed selection could lead to reduced yield and economic loss. To tackle this problem, this study proposes an automated machine learning based classification of wheat seed varieties. The experiments are carried out on six wheat varieties grown in Southern Punjab, Pakistan. This research utilizes the data obtained by taking digital images of wheat seeds under consistent lighting and a uniform background at three different resolutions: 256 x 256, 512 x 512, and 1024 x 1024. In seed analysis, texture is chosen to describe seed surface characteristics because texture information is best at differentiating visually similar seeds. These images are used to calculate the following features: Histogram Features, Texture Features, and Laws Texture Features using Spatial Gray Level Co-occurrence Matrices. These matrices are second-order statistics such as correlation, energy, contrast, and homogeneity calculated based on the relationship between the intensities of the pixels at various distances and orientations. Then, the derived features are optimized using Principal Component Analysis and Correlation Feature Selection. Five machine learning classifiers, namely LogitBoost (LB), BayesNet (BN), Classification Via Regression (CVR), Multi-Layer Perceptron (MLP), and Random Forest (RF), are employed. The classifiers are evaluated separately on the PCA-reduced and CFS-selected datasets for each image resolution. Classification accuracy and consistency were assessed. The experimental results indicated that the PCA optimized Dataset of 1024 x 1024 resolution and the Random Forest classifier gave excellent discrimination between the wheat varieties, and the classifiers achieved 96.4% accuracy for all the classes. The proposed system helps to minimize human judgment dependency and ultimately reduces the risk of misclassification at the early stage of cultivation. The method is non-destructive, economical. and applicable in agricultural applications.

@article{Ahmed2026,
  title     = {A Machine Learning Framework for Wheat Seed Variety Classification Using GLCM-Based Texture Analysis and Feature Optimization},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {9},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Muhammad Munawar Ahmed and Malik Muhammad Saad Missen and Hannan Adeel and Ahamad Zaki Bin Mohamed Noor and Muzamil Malik},
  doi       = {10.14569/IJACSA.2026.0170939},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170939}
}