Knowledge Extraction through Floweret Recognition using ML and DS: A Sustainable Approach for 21st century Agriculture and Smart Cities

Recognizing flowers presents a formidable challenge due to the considerable similarity among various species in terms of size, shape, color, and the presence of surrounding elements like leaves, grass, petals, sepals, and stems. In this study, the authors propose an innovative two-stage deep learning classifier aimed at distinguishing between various species of flowers. Initially, an automated blossom segmentation process is employed to isolate the flower region, facilitating the creation of a minimal bounding box around it. This segmentation step is integral for narrowing the focus to the relevant flower area and minimizing the impact of extraneous visual information. Subsequently, the authors develop a robust convolutional neural network (CNN) classifier tailored for different types of flowers. The CNN is designed to effectively capture and differentiate the diverse characteristics of various flower species. The combination of these two stages—segmentation and classification—provides a comprehensive approach to enhance the accuracy of flower recognition, particularly in cases where species exhibit similarities in size, shape, and color. To better define flower recognition, our team first applied CNN (Convolutional Neural Network) with one, three and four convolutional layer and then ResNet50 pre-trained model in which we found fixed and best accuracy in comparison to others. From CNN with four-convolution layer we attain the accuracy of 97.1%