Curvature-based pattern recognition for cultivar classification of Anthurium flowers

Real-time classification of agricultural products with various cultivars is an important issue in postharvest processing, which speeds up the processing and consumer delivery time. An innovative approach was developed for cultivar classification of Anthurium flowers based on image processing, B-spline curves, mathematical operations and machine learning classifiers. The algorithm was implemented and tested on a database of Anthurium flower images, which included the images of 15 cultivars of the flower with various sizes and shape categories. The boundary of the flowers was detected and reconstructed using a suitable B-spline curve. The signed curvature of the curve was calculated via mathematical operations. Then, several classifiers were implemented using the machine learning methods, Support Vector Machines (SVM), K-Nearest Neighbors, Discriminant Analysis, Decision Trees, and Naive Bayes, to detect and classify the cultivars of the flower. The experiments were carried out using a different number of training samples of the database images. The effect of various classification methods and variations in the angle of rotation of placing the flowers under the camera on classification accuracy were evaluated and the computation time of the classification process was measured. The results showed that in the unrotated sample with 1.5 pixels/mm density, the classification accuracy of the Naive Bayes and SVM algorithms had the highest classification accuracies, more than 98%. Also, the Decision Trees classifier had the lowest computation time, less than 2.5 ms. The proposed approach had proper classification accuracy and low computational load, which could be used in the real-time classification systems for Anthurium flowers.