Cauliflower Disease Recognition Using Machine Learning and Transfer Learning

Abstract

In terms of overall winter cropping area and production in Bangladesh, cauliflower dominates a large share. It has many health benefits like decrease the risk of obesity, diabetes, and heart disease. It is a cultivated and winter crop which has huge demand in the country. But if proper care is not taken, many serious diseases will affect plants and will reduce productivity, quantity, and quality of cauliflower. Manual monitoring of plant disease is very difficult as it requires a tremendous amount of work and excessive time. Automatic recognition of disease through computer vision approach is becoming more popular day by day. So, in this paper, we introduced several techniques to recognize diseases that occur on plants in cauliflower. Our proposed solution would support the agriculture field of Bangladesh to grow cauliflower more effectively and will increase its production by taking the proper step after automated recognition of diseases. In our work, we have compared traditional machine learning and transfer learning. In machine learning, for image segmentation, k-means clustering is used after the image preprocessing method is applied, and then, ten relevant features are extracted. For classification, we compared various classification techniques. Random forest algorithm achieves overall 81.68% accuracy. Different CNN-based architectures with transfer learning, namely InceptionV3, MobileNetV2, ResNet50, and VGG16, are also applied. InceptionV3 achieves 90.08% accuracy which is the highest accuracy among these two approaches.