A novel method to improve computational and classification performance of rice plant disease identification

Rice is a major food crop that plays an important role in the Indian economy. It is the most consumed staple food, greatly in demand in the market to meet the requirements of a growing population, which is only possible with increased production. To meet this demand, rice production should be increased. To maximize crop productivity, measures must be taken to eradicate rice plant diseases, namely, brown spot, bacterial leaf blight, and rice blast. In the proposed method, the modified K-means segmentation algorithm is used to separate the targeted region from the background of the rice plant image. Following segmentation, features are extracted through the three parameters of color, shape and texture. A novel intensity-based color feature extraction (NIBCFE) proposed method is used to extract color features, while the texture features are identified from the gray-level cooccurrence matrix (GLCM) and bit pattern features (BPF), and the shape features are extracted by finding the area and diameter of the infected portions. Thereafter, unique feature values are identified through the novel support vector machine-based probabilistic neural network (NSVMBPNN) to classify the images. A comparison in terms of performance is made using three classifiers, namely naive Bayes, support vector machine and probabilistic neural network. This proposed method achieved better accuracy than the other three methods based on different performance measures. Finally, the result was validated under the fivefold cross-validation method with final accuracies of 95.20%, 97.60%, 99.20% and 98.40% for bacterial leaf blight, brown spot, healthy leaves and rice blast, respectively.

Automated summary

Rice is a major food crop in India and increasing its production is crucial to meet the demands of a growing population. This study proposes a novel method to identify rice plant diseases through segmentation, feature extraction, and image classification. The results show better accuracy compared to other methods.