[Paper Review] Flower Categorization using Deep Convolutional Neural Networks

Ayesha Gurnani, Viraj Mavani+2 more|arXiv (Cornell University)|Aug 12, 2017

Smart Agriculture and AI8 references21 citations

TL;DR

This paper proposes a deep learning approach for flower categorization using two CNN architectures—GoogLeNet and AlexNet—on the Oxford 102-Flower dataset. It achieves top-1 and top-5 accuracies of 47.15% and 69.17% with GoogLeNet, significantly outperforming random classification (0.98%), demonstrating strong performance for fine-grained flower recognition in real-world applications such as botany and nature exploration.

ABSTRACT

We have developed a deep learning network for classification of different flowers. For this, we have used Visual Geometry Group's 102 category flower dataset having 8189 images of 102 different flowers from University of Oxford. The method is basically divided into two parts; Image segmentation and classification. We have compared the performance of two different Convolutional Neural Network architectures GoogLeNet and AlexNet for classification purpose. By keeping the hyper parameters same for both architectures, we have found that the top 1 and top 5 accuracies of GoogLeNet are 47.15% and 69.17% respectively whereas the top 1 and top 5 accuracies of AlexNet are 43.39% and 68.68% respectively. These results are extremely good when compared to random classification accuracy of 0.98%. This method for classification of flowers can be implemented in real time applications and can be used to help botanists for their research as well as camping enthusiasts.

Motivation & Objective

• To develop a deep learning-based system for accurate classification of flower species from images.
• To evaluate and compare the performance of two established CNN architectures—GoogLeNet and AlexNet—on a fine-grained flower classification task.
• To enable real-time flower categorization for practical applications in botany and nature-based research.
• To provide a robust, scalable solution for automated flower recognition that supports both academic and field-based use cases.

Proposed method

• Utilized the Oxford 102-Flower dataset, consisting of 8,189 images across 102 flower categories.
• Applied a two-stage pipeline: image segmentation followed by deep learning-based classification.
• Trained and evaluated GoogLeNet and AlexNet models with identical hyperparameters for fair comparison.
• Employed standard deep learning training protocols, including data augmentation and stochastic gradient descent.
• Used top-1 and top-5 accuracy as evaluation metrics to assess model performance.
• Ensured consistent training and validation splits to maintain reproducibility and reliability of results.

Experimental results

Research questions

• RQ1 How do GoogLeNet and AlexNet perform in classifying fine-grained flower species from natural images?
• RQ2 Can a deep CNN-based system achieve meaningful accuracy on a challenging, fine-grained flower classification benchmark?
• RQ3 What is the comparative performance of GoogLeNet versus AlexNet in terms of top-1 and top-5 classification accuracy on the 102-Flower dataset?
• RQ4 To what extent does the proposed method surpass random classification in flower categorization tasks?
• RQ5 Can this system be adapted for real-time deployment in field applications such as botany or nature exploration?

Key findings

• GoogLeNet achieved a top-1 accuracy of 47.15% on the 102-Flower dataset, significantly exceeding random classification accuracy of 0.98%.
• AlexNet achieved a top-1 accuracy of 43.39%, demonstrating strong but slightly lower performance than GoogLeNet.
• GoogLeNet achieved a top-5 accuracy of 69.17%, compared to 68.68% for AlexNet, indicating better class ranking capability.
• Both models outperformed random classification by a substantial margin, validating the effectiveness of deep CNNs for fine-grained flower recognition.
• The consistent performance gap between GoogLeNet and AlexNet suggests architectural advantages in GoogLeNet for this specific task.
• The results indicate that the proposed method is suitable for real-time applications in botany and nature-based research.