A Methodology of Customized Dataset for Cotton Disease Detection Using Deep Learning Algorithms

Agribusiness occupies the lion’s share of Pakistan’s land. It also supports Pakistan’s financial situation. Approximately 62% of Pakistan’s population lives in rural regions and relies on agriculture for a portion of their income. Pakistan is now the 5th-largest producer of cotton and the 3rd-largest consumer/manufacturer of cotton yarn worldwide. Cotton is grown on 6.0 million acres by 1.3 million of the country’s 5 million farmers, or around 15% of the country’s total cultivated land. Cotton fields are plagued by various illnesses that may have a devastating effect on the quality and quantity of the crop. Detection of these disorders has become more common because of image processing. Pathogens often cause plant diseases like germs, fungi, and microbes that thrive in an unsanitary environment. The farmer suffers a significant setback as a result of this. The main purpose of this research is to get to know the disease in a cotton field. We identify plant/leaf disease using the convolutional neural networks (CN) technique with Pooling, Flatten, Dense, and dropout layers to analyze picture data using TensorFlow and Kera’s support. Our dataset has six classes, including 1965 photos of five sick cotton plant classes and one healthy class. We compared three Kera’s applications to get the algorithm’s best accuracy. The applications we used are Xception, InceptionV3 and InceptionRestNetV2. The Xception model shows us the best accuracy, an average of 90.34%.