Flower Image Classification

Flower Classification Diagram

Overview

This project builds an image classification system to identify 102 flower species using a pretrained MobileNetV2 model. Transfer learning is applied to efficiently adapt the model to the dataset.

The model is trained on the Oxford Flowers 102 dataset and performs predictions on unseen images.

Goal: Train a deep learning model for multi-class flower classification using transfer learning.
Notebook: View Code

Key Highlights

  • Transfer learning using MobileNetV2
  • Fine-tuned for 102 flower categories
  • Lightweight and efficient architecture
  • End-to-end pipeline from training to inference

Dataset

The dataset contains 102 flower categories with variations in scale, pose, and lighting.

  • Training images: 1020
  • Validation images: 1020
  • Test images: 6149

Images are resized to 224 × 224 and normalized using ImageNet statistics.

Model

A pretrained MobileNetV2 model is used as the base architecture. The classifier layer is replaced to match the 102 classes, and transfer learning with partial fine-tuning is applied.

  • Total parameters: 3,504,872
  • Trainable parameters: 130,662

Results

Sample predictions from the trained model