Most search engines retrieve photographs using classic text-based algorithms that depend on descriptions and metadata. Content-based image retrieval (CBIR), image categorization, and investigation have all gotten a lot of attention in the previous two decades. High-level picture views are represented as feature vectors consisting of numerical values in CBIR and image classification-based algorithms. To isolate the main object from a picture, we first use segmentation and main object detection. The autoencoder is then used to extract features from the object and choose relevant features. Various deep learning representations are trained and tested, and the outcomes are compared to see which architecture maximises prediction scores while reducing computing costs in flower categorization identification. Images of flowers will be used to train and test models, with a selection of them being utilised for validation. Finally, the results of the experiments suggest that our technique can be used to search for images in a genuine picture database.