Development of forecast models using weather and satellite based agromet products for crop yields, pests & diseases of target crops. The methodologies used for development of forecast models are

Developed and validate the forecast models (statistical as well as machine learning approach) for yield & major pests and diseases of Rice, Wheat, Cotton, Mustard, Chickpea and Pigeonpea at various locations using Agromet and SATMET data on various character viz. (i) Crop age at first appearance of pests (ii) Crop age at maximum population of pests (iii) Maximum pest population and (iv) Weekly monitoring of pests. The developed models were converted into web-based for systems on the basis of three-tier architecture, which includes Client Side Interface Layer (CSIL), Application Logic Layer (ALL) and Database Layer (DBL) for operational purpose.

Image Analysis based on CNN (Convolution Neural Network) for target diseases in target crop with ReLU activation function has been done. In CNN (Convolution Neural Network) architecture, feature maps from previous layers are convolved with learnable kernels. The output of the kernels goes through a linear or non-linear activation function (sigmoid and hyperbolic tangent) to form the output feature maps. In this image analysis, ResNet is a traditional feedforward network with a residual connection were attempted, in this network the output of a residual layer can be defined based on the outputs of pervious layer performing various operations such as convolution with different size of filters and Batch Normalization (BN).

Dr. Amrender Kumar, Incharge & Principal Scientist
Ms. Ritika, Scientist
Mr. Ashish Sharma, STO
Ms. Rakhi Sharma, RA
Ms. Shelly, SRF