Comparison of Image Features Descriptions for Diagnosis of Leaf Diseases

The agricultural industries have always demanded technologies for the automatic discovery and diagnosis of plant diseases with high speed, accuracy, and low cost. Numerous studies have been conducted in response to this demand; however, significant issues remain in most cases where a large-scale dataset of field images is taken with different atmospheric conditions, lighting, scale, and in different directions. The large dataset often causes high computational and storage costs. To overcome this problem, we focus on methods based on efficient invariant image features. These methods are robust against such external factors added during image acquisitions with low computational cost and higher accuracy. We then use a well-known data clustering algorithm k-means to create visual features for lesions. We then create a group of robust visual features (BoVF) using the Term Frequency-Inverse Document Frequency (TF-IDF) weighting scheme that considers the most important visual features in the image for classification. Experimental results classify the BoVF using K-means clustering that categorizes a particular disease in the leaf image into their appropriate group.