Identification of Local Flowers Through Image Processing

Abstract

Flowers are one of the most beautiful creation of almighty. Bangladesh is the land of flowers. There are lots of flower in our country. There are many different kinds of flower in our country. Every flowers have its own color, shape and name. There are many seasonal flowers in this country. Now a days flowers are used in every occasion like wedding, birthday party etc. In our everyday life, we can see many different kinds of flowers around us while we walking beside roads, rail line even in our garden. But most of our people don’t have any knowledge about that flowers. Even they don’t know the flowers name. But many of them wants to know about the flowers. For that reason we choose this topic to research and develop our system. Through our system people will be known about that unknown flowers which they see but don’t know about that flowers. Our system based on neural networks to create an image classification by Tensorflow. CNN is the most popular platform of machine learning and it is extensively used for image classification. We used some CNN models like Inception-v3, VGG-19, MobileNet and ResNet-50 for detecting and classifying flowers. From all of the model we got best accuracy (98.76%) by using MobileNet model. We have collected six types of flower for our project. Each flower have 400 images and in total 2400 image in our dataset. We kept 60% image in training set for train the dataset. 20% kept in validation set and another 20% in test set for testing purpose. In this system we use six types of flower but in future we have plan to add more flower for upgrading our system