Plant-Wise Comparison of Various CNN Algorithms for Detection of Pest Infestation

Agriculture is the primary sector of income in countries like India. Pest attacks to crops are a major concern in these countries. The symptoms of pest infestation can appear on various parts of the crop such as leaves, stem, roots, etc. This paper focuses on comparing the different Convolutional Neural Network models to detect pest infestation. The images of crop leaves (both diseased and healthy) have been considered). Various crops such as potato, grapes, cherry, bell peppers, etc. with diseases such as Bell pepper - Bacterial spot, Cherry - Powdery Mildew, Potato - Early Blight, etc. have been considered to arrive at an unbiased conclusion. For each model different optimizers such as Stochastic Gradient Descent, ADAM, etc have been implemented. Batch normalization has been applied to each of the models. Among the CNN architectures, Alexnet, ResNet-50, Inception V3, and VGG-16, ResNet-50 gives a higher accuracy for most of the plants ranging between 95.99% for grapes, 99.34% for both potato and bell pepper. Inception V3 also yielded good results for a few plants, with accuracies ranging between 99% for strawberry and 99.75% for cherry. These models with the aforementioned parameters have been implemented for a low computationally efficient device.