Comparative study on the effect of activation functions in neural network for rubber tree diseases detection

• Main Author: Mohd. Rasidi, Haryani
• Format: Thesis
• Language: English
• Published: 2011
• Subjects: Q Science (General)
• Online Access: http://eprints.utm.my/id/eprint/32825/5/HaryaniMohdRasidiMFSKSM2011.pdf; http://eprints.utm.my/id/eprint/32825/

The rubber tree (Hevea brasiliensis) is grown extensively in South-East Asia, especially in Malaysia, for the production of natural rubber and, increasingly, timber. However, rubber tree generally attacked by pests or diseases such as root disease, white root, red root, brown root, leaf disease and the others. In this study, four types of diseases will be seen namely Fusicoccum, Corynespora, Collelotrichum and Oidium. Generally known that neural networks may be used for nonlinear analysis of complex data. As such, the purpose of this study is to evaluate the usefulness of Artificial Neural Networks (ANNs) applied to rubber tree for diseases detection. The ANNs used in the present study were based on a feed forward layered model with input, hidden, and output layers, on which a backpropagation learning model was implemented. The focus of study also to investigate the effect of activation functions on accuracy, efficiency and performance of disease detection. Four activation functions will be compare which are Hyperbolic Tangent Sigmoid, Linear, Radial Basis and Triangular. This study also will be focus on how to convert an image data to the conventional input data. The preprocessing process of the sample image such as image enhancement, image filtering has been performed before the process feature extraction will be applied. The techniques that will be used for feature extraction of image using the gabor filter method. The output of this filter will be used for the classification purpose in order to determine the characteristics of rubber tree diseases image. 10-fold cross validation techniques will be applied in order to measure the percentage of accurately of this classifier.