BloomNet: An eco-aware lightweight deep learning for medicinal flower recognition

AbstractFlower recognition plays a vital role in connecting human and nature. In plant species identification, deep learning has surpassed many algorithms, ranging from complex, high-accuracy systems to lightweight architectures. However, Indian medicinal flower-based plant recognition system remain underexplored in this AI era. Recognizing flowers despite appearance (color, texture and shape) and image (rotation, illumination and viewpoints) based challenges are tedious. In this paper, flower images are preprocessed. A customized Convolutional Neural Network (CNN) called “BloomNet” is constructed using the pre- processed images as input. The custom-developed BloomNet contains five layers including convolution and fully connected layers. BloomNet is evaluated on two datasets, namely, custom-collected Med-3Flower and Flowers-17. The proposed methodology produced an accuracy of 100% for Med-3Flower and 87.65% for flower-17 datasets.