Machine learning approach for the classification of wheat grains

Wheat is known to be one of the most important agricultural crops throughout the world. Due to its mass storage in warehouses, tonnes of wheat grains rotten every year that eventually affects its market price. This paper presents an end-to-end automatic system that utilizes computer vision techniques for the quality grading of wheat grains. The main purpose of this work is to determine most discriminatory features and a suitable classifier that may classify the given wheat sample into two classes 'fresh' and 'rotten'. At first, shadow removal, segmentation, and separation of each grain is performed as parts of the pre-processing step. The pre-processing is followed by the features extraction step where 7 color and 16 texture handcrafted features are determined for each grain. The four binary classification models, namely, Support Vector Machine (SVM), K-Nearest Neighbour (KNN), Multi-Layer Perceptron (MLP), and Naïve Bayes (NB) are then built using 10-fold cross validation approach. The classifiers are compared on the basis of performance metrics- accuracy, error rate, recall, specificity, precision, and F1-score. The comparative analysis depicts that based on color features, the SVM classifier outperforms other classifiers by achieving the accuracy of 93%. In contrast, based on texture features, the NB classifier achieved accuracy at 65%; highest among all classifiers. Experimental results encourage the utility of SVM classifier modelled on color features in automatic quality grading of wheat grains.