Machine Learning Approach For The Classification of Sunflower Seed Varieties Using Hybrid Features

Abstract

Sunflowers are a vital crop for modern agricultural industry. Worldwide, sunflower production is estimated to be 23 million hectares in 60 countries. The grower’s concerns regarding the originality of the sunflower seed variety and quality. The purity of seed is an essential quality indicator for crop seed. The goal of this research was just to investigate feasibility of a ML (machine learning) technique for identifying various kinds of sunflower seed varieties. The DI (digital images) of 6 sunflower seed varieties were Aguara-4, Armoni, T-40318, Fh-675, Fh-701, US-444. This was accomplished using a digital camera in a naturalistic environment even without a specialized laboratory system. The obtained digital picture collection is transformed into a hybrid feature dataset, which combines histogram, texture, binary, and spectral features. On each non-overlapping region of interest, sizes (32 × 32) total number of 55 hybrid features were obtained for every sunflower seed digital images. Three optimal features were obtained by using the correlation-based feature selection technique (CfsSubsetEval) with the BFS (BestFirstSearch) algorithm. To develop the classifications algorithms, J48, RandomTree, Random committee, Bayes net, and Bagging were used with an optimized multi feature cross validation technique (10-fold). A comparative study of five Machine learning classifiere, J48 performed well in classification accuracies are (99.9729%), when region of Interests size (32 × 32) within 0.01seconds.