Paphiopedilum Orchid Flower Classificatiom Using A Multilayer Perceptron Neural Network

Classification is an important task in digital image processing. In the field of automatic flower classification research there has been great interest in this topic for many years. Recently, where groups can be recognized, researchers have attempted to classify large.scale flower species. In the case of unknown species’ names, classification of flower images can be performed based on visual content. Therefore, flower classification is an interesting problem and a challenge for the community. The main problem facing all researchers involving the with flower classification systems is to avoid being affected by different variations in the appearance of the flowers such as the color, shape, size and texture. Additionally, in a natural setting, images of flowers are often taken in outdoor scenes where the lighting condition varies with the weather and time. However, a robust classification scheme is needed to help the researcher find the scientific names of the flowers, conveniently and easily. It also helps botanists and flower specialists identify suspect species precisely. Moreover, it can reduce the duration of work and errors caused by humans.In this thesis, we aim to develop a model to classify the Paphiopedilum Orchid Flowers. We select some of the most famous types of orchids in Thailand. They are colorful, an endangered plant species and also quite interesting because they tend to produce only a single blossom per plant. Additionally, there are many species with a similar appearance, which makes the task of classification, difficult and laborious. The thesis focuses on how to classify Paphiopedilum Orchid Flowers, we used a Multilayer Perceptron Neural Network (MLP) model to classify visual content of the flower images, features were extracted from Color and Segmentation-based Fractal Texture Analysis (SFTA) features as input feature vectors for classification. The main contributions of this thesis are summarized as follows:(1) We have created our own database of Paphiopedilum Orchid Flowers, in which 1100 images of 11 Paphiopedilum Orchid Flowers species are included in our dataset, it has more variety of species than any previous works (they use 200 images of 5 Paphiopedilum Orchid Flowers species). Also, although a lot of research has been done on flower classification of large-scale flower species, Paphiopedilum Orchid Flowers have rarely been studied before in China. Therefore, Paphiopedilum Orchid Flowers are still in need of research for flower classifications.(2) We propose a new approach for Paphiopedilum Orchid Classification (POC) using Multilayer Perceptron Neural Network (MLP) classifier which achieves high classification accuracy in our experiments. Compared with previous works, our model more efficiently classifies Paphiopedilum Orchid Flowers and is easily extended to other similar classification tasks. We applied a variety of feature extractions. For color features, we implemented two types of color feature extraction algorithms:Color Moments and Color Histogram. Additionally, we applied many popular color models such as RGB, CIE XYZ, YCbCr and HSV in our experiments. HSV color is the best color feature for this thesis experiment. Moreover, the texture feature extraction we used is Segmentation-based Fractal Texture Analysis (SFTA) algorithm. These feature extraction algorithms are of great importance to the quality and efficiency and improve our classification performance better than previous works. Furthermore many approaches of previous works use a variety of classifiers such as a random forest, Support Vector Machine (SVM), Artificial Neural Network (ANN) classifier, etc. Therefore, for this research we also implemented several renowned classifiers such as Navie Baye, K.nearest, Binary tree, Sequential Minimal Optimization (SMO) and Multilayer Perceptron Neural Network in our experiments. Our work shows that the most robust classifier is Multilayer Perceptron Neural Network which yielded highest accuracy 97.64%.As for the results of this dissertation; our Paphiopedilum Orchid Classification model give us accurate results to a satisfactory level. Additionally, our model can be applied to a flower image retrieval method and also will help classify different types of flowers in the future.