Please use this identifier to cite or link to this item: http://hdl.handle.net/10266/3761
Title: Plant leaf image classification using artificial neural networks
Authors: Singla, Naveen
Supervisor: Sachdeva, Jainy
Keywords: ANN;Computer aided classification (CAC);Hu Invariant;electrical;electrical and instrumentation;EIC;EIED
Issue Date: 7-Sep-2015
Abstract: In this dissertation, computer aided classification (CAC) system has been proposed to classify maple species plants. These maple species are important to classify as some species are helpful in maintaining ecological balance and the others have high medicinal values. The five maple species namely Acer monspessulanum (AM-Class-1), Acer negundo (AN-Class-2), Acer opalus (AO-Class-3), Acer Campestre (AC-Class-4) and Aesculus hippocastanum (AH-Class-5) are taken from ImageCLEF 2012 database. The total 155 intensity and texture features extracted are first order statistics (FOS), gray level co-occurrence matrix (GLCM), Gabor filter, Gabor wavelets and shape based Hu moments. Artificial neural network (ANN) is used to classify maple plant species based on the features extracted. Two sets of experiments are performed. In the first experiment 68 features are taken. It is observed that individual classification accuracy for AM (Class-1) is 83.33%, AN (Class-2) is 82.14%, AO (Class-3) is 86.37%, AC (Class-4) is 93.26% and AH (Class-5) is 95.45% and overall classification accuracy is 78.98%. In the second experiment strong shape and texture based features such as Hu moments and Gabor wavelets are specifically added in the feature bank. It is observed from the results that higher overall classification accuracy is achieved i.e. 91.08%. The better individual classification accuracy is also achieved for every class i.e. for AM (Class-1) is 91.67%, AN (Class-2) is 92.86%, AO(Class-3) is 86.37%, AC (Class-4) is 87.55% and AH (Class-5) is 100%.
Description: ME-EIC-Thesis
URI: http://hdl.handle.net/10266/3761
Appears in Collections:Masters Theses@EIED

Files in This Item:
File Description SizeFormat 
3761.pdf1.72 MBAdobe PDFView/Open    Request a copy


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.