CNN fruits-360

Project Description: CNN Improved Fruits-360 Classification

This project focuses on developing an enhanced Convolutional Neural Network (CNN) model for image classification using the Fruits-360 dataset. The dataset contains thousands of labeled fruit and vegetable images divided into 219 classes, making it a challenging multi-class classification problem.

The main objective was to build a deep learning model capable of accurately recognizing different fruit categories from images. Data preprocessing techniques such as rescaling, rotation, zooming, shifting, and horizontal flipping were applied using ImageDataGenerator to improve generalization and reduce overfitting.

The CNN architecture was designed with multiple convolutional blocks, including:

Conv2D layers for feature extraction

BatchNormalization for training stability

MaxPooling2D for dimensionality reduction

Dropout layers to prevent overfitting

Fully connected Dense layers for final classification

To optimize performance, the project used:

EarlyStopping to stop training when validation loss stopped improving

ReduceLROnPlateau to lower learning rate automatically

ModelCheckpoint to save the best model during training

Final Results:

Training Accuracy: 98.03%

Validation Accuracy: 93.56%

Test Accuracy: 97.20%