Deep learning model such as AlexNet, VGG, ResNet, Inception, DenseNet requires large parameter to train the model. Implement these models in agriculture field requires high powered devices which may not be feasible in agriculture domain. In this point of view a shallow CNN network and hybrid CNN model is proposed which uses very much less parameter as compared with the different well known DL models. In one approach, identification of the plant diseases using shallow VGG network is proposed which uses only seven layers of VGG19 model. In another approach, a CNN which uses inception layer with residual connection is build. The number of parameters used in this model is much less as compared with several models