Identification of Adulterated Wheat Varieties Using Optimized Machine Learning Algorithms

Abstract

Adulteration in crops widely affects the production, despite maximization of yield in agriculture industry. In the traditional system, farmers discriminate wheat grains of various varieties based on visual inspection, which is tedious task. Therefore, automatic approaches for identification of adulterant wheat varieties essential. This led to take cost-effective technologies like Computer Vision (CV) and Machine Learning (ML) for the detection of adulterated wheat varieties from bulk samples. The study presented machine learning based classifiers using Back Propagation Neural Network (BPNN) and Support Vector Machine (SVM) in classifying seven adulterated wheat varieties across five adulteration percentages of 10%, 15%, 20%, 25% and 30%. Combined color and texture features are extracted from the image samples and auto-encoder decoder method is applied to facilitate feature selection. To evaluate the obtained classification accuracies, a Deep Learning (DL) approach using hybrid CNN-SVM classifier is employed. The study compares performance metrics of the models tested in classification accuracy and computational time. The proposed non-destructive automatic methods are helpful for the testing quality and authenticity of wheat grain samples.