Wheat is a major crop in Pakistan’s agriculture-based economy and a primary source of food for the population. Accurate classification of wheat varieties is essential for improving crop productivity and supporting breeding programs. This study focused on six high-yield hybrid wheat varieties: BLA4, Chenab Pasta-24, H1, H1A, H2B, and Arooj-22. Traditional identification methods rely on seed characteristics, which are time-consuming and ignore other important morphological features such as the wheat spike. A Spike-Vision-based classification method was developed using machine vision and machine learning. A total of 1,500 spike images were acquired, with 250 images per variety, using a smart camera under controlled conditions. The images were preprocessed to improve quality and consistency, then segmented using HSV thresholding to isolate spike regions. MobileNetV2 was used for feature extraction to capture morphological details of the spikes. Multiple machine learning models were tested for variety classification. Random Forest achieved the highest accuracy of 98.91%, showing strong performance in distinguishing varieties based on spike features. This model effectively handled complex variations in spike morphology and outperformed other classifiers in the study. The results confirm that spike-based classification using image analysis and machine learning provides a fast, accurate, and scalable alternative to conventional seedbased methods. This approach reduces manual effort, enhances identification precision, and supports modern agricultural applications such as crop monitoring, precision breeding, and automated phenotyping.
Loading....