CFNN for Identifying Poisonous Plants

Abstract

Identification of poisonous plants is a hard challenge for researchers because of the great similarity between poisonous and non- poisonous plants. Traditional methods to identify poisonous plant can be tiresome, therefore, automated poisonous plants identification system is needed. In this work, cascade forward neural network framework is proposed to identify poisonous plants based on their leaves. The proposed system was evaluated on both (poisonous leaves/non-poisonous leaves) which are collected using smart phone and internet. Combination of shape features and statistical features are extracted from leaf then fed to cascade-forward neural network which used TRAINLM function for training. 500 samples of leaf images are used, 250 samples are poisonous, the remaining 250 samples are non-poisonous.300 samples used in training, 200 samples for testing. Our system is achieved an accuracy value of 99.5%.

Keywords

• Cascade Forward Neural Network (CFNN), First Order Statistical Features, Poisonous Plants, Shape Features, TRAINLM Function