ANALYSIS OF ACTIVATION FUNCTION IMPACT ON DEEP LEARNING IN CASE OF POTATO LEAF DISEASE DETECTION

For many years, potatoes have been staple human diets, originating in the southern part of North America and subsequently spreading across the world. A global food system will need to be significantly improved if it is to be able to sustainably and nutritiously feed the growing world population in the next decades. Detecting potato leaf disease is essential in the agricultural sector to ensure food security. Deep learning techniques have demonstrated impressive performance in detecting potato leaf disease. The objective of this study is to identify the best activation function to effectively detect infections in potato leaves. Potato leaf images were collected from Masha wereda gatimoy kebele South West Ethiopia using mobile camera. Total 2000 potato leaf images were taken from Masha and Gecha farm area and Masha Agricultural Fields. These have three channels Red (R), Green (G) and Blue (B). On this study image preprocessing is performed by resizing the image to 128 * 128 pixels and converting to gray scale. Finally, the proposed CNN model was evaluated using performance evaluation metrics such as accuracy, precision, recall and F-Score. Based on the findings of the experiments, the leaky rectifier function achieved 98.0%accuracy. Therefore, LRelu activation function is best activation function for the convolutional neural network (CNN) model. The study utilizes deep learning techniques to show the identification of potato leaf diseases, aiming to offer valuable insights to researchers and practitioners in the agricultural field. By employing convolutional neural networks (CNN) in agriculture, the intention is to enhance potato production by providing heuristic knowledge