VADODARA: In an effort to stem crop damage and potentially a crucial step towards food security, artificial intelligence (AI) and machine learning (ML) techniques are being used to gather knowledge about pests and diseases ruining cereals like rice, maize, wheat and sorghum. Under a planning scheme funded by the Gujarat government under the Department of Basic Sciences and Humanities, Anand Agricultural University (AAU) has begun research work in this direction. Around 30 per cent of the annual crop yield in India is wasted due to pests. According to a report by the Indian Council of Agricultural Research, 60 million tonnes of crop is lost annually due to pests and nematodes. The aim is to study the impact of diseases and pests on grain quality and diagnose such problems using various computer vision models. Data for the project will be collected from leading research centres like the Key Rice Research Centre at Navagam, Key Maize Research Centre at Godhra. The research centres include the Agricultural Research Station (Wheat) at Arnej and Key Sorghum Research Station at Surat. Along with the research centres, scientists will also collect information and images of crop pests and diseases from farmers' fields near the centres. “The results of this study will help farmers pinpoint and proactively protect against diseases and pests affecting these important cereal crops. Farmers will be trained and guided in this,” said Dr KB Kathilia, vice-chancellor, AAU, who is directing the project along with Director of Research Dr MK Jhala and Principal and Dean Dr YM Shukla. The project will prepare image data by taking precise pictures of pests and diseases that cause damage according to the season and different stages of the crop. Image processing will then be done on these pictures. The image output will further be provided as input to machine learning (ML) techniques such as neural networks and deep learning models for model training and analysis. “Images of pests and diseases of crops will be generated using high performance cameras and an image database will be created,” he said. Data processing will be carried out using support vector machines, artificial neural networks, nearest neighbour neural networks, probabilistic neural networks, backpropagation neural networks, etc. Each model is expected to deliver results. “These models have reported accuracy rates ranging from 75 to 99.99 percent, which means that diagnosing pests using the models is 75 to 99.99 percent accurate,” the official said.
