Sensitivity Analysis of Coffee Leaf Rust Disease using Three Deep Learning Algorithms

Abstract

Coffee is the second most traded commodity in the world. The global coffee production is mainly affected by Coffee leaf rust disease, which can cause production loss of up to 35%. To combat the disease, a long- term solution in the form of early detection of the coffee leaf rust in coffee plants is required. This paper applies three deep learning algorithms for the detection of coffee leaf rust infection rate, namely Backpropagation Neural Network (BPNN), Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN). The modified algorithms are executed using TensorFlow framework on the dataset obtained from Colombia and evaluated using Mean Absolute Error (MAE). Each algorithm was executed while varying sensitivity and perturbation parameters. The sensitivity parameters used are: the type of activation function and the number of input parameters while the perturbation parameters used are: the number of epochs and the data division ratio. The results show that the best recorded algorithm is the modified BPNN algorithm which yields a minimum MAE of 1.2462 after 1500 training epochs when the dataset was subdivided into 50:50 ratio. The results can be applied to develop an early detection system that will be used by farmers to prevent their coffee plants from further damage.