Advanced Deep Learning techniques for Cauliflower Plant Disease Detection

The agriculture sector in India is growing and the field of technology is contributing significantly to this sector. Researchers are constantly trying to find ways to use artificial intelligence which can aid this field. Disease classification in crops using deep learning models is an active area of research currently. The work presented in this paper uses cauliflower (Brassica oleracea var. botrytis) as the target crop for disease detection as it is one of the most important crops in India and is easily prone to diseases. Despite the growth of AI in agriculture world and research on crop disease detection, most of the studies focus mainly on crops like wheat and rice, leaving the specific challenges of cauliflower largely unexplored. Cauliflower holds substantial economic importance in India. India is the second largest producer of cauliflower globally after China. India ranks second in area and production of cauliflower in the world after China. In India, Cauliflower is grown with production of 8.668 million tonnes. The crop is grown in various states of India, including West Bengal, Punjab, Haryana and Uttar Pradesh and generates significant income for farmers. However, diseases like bacterial rot spot, downy mildew, and black rot can lead to major yield losses, affecting farmers’ incomes and the broader economy.We have applied eight deep learning models which are Xception, EfficientNet B0, EfficientNet B1, EfficientNet B2, EfficientNet B3, EfficientNet B4, ResNet15 v2, DenseNet201 which were evaluated using accuracy, loss, recall and precision. Out of all these models, EfficientNet B3 performed the best with an accuracy of 99.85% and a loss of 0.0023. This model was closely followed by EfficientNet B0 and EfficientNet B2, they had the same accuracy, but the value of loss was 0.0054 and 0.0055 respectively. The findings suggest the potential of deep learning techniques for improving disease detection in cauliflower.