Implement CNN from scratch for celestial classification

[yogesh4216]

Issue: #134

Model Description

A Convolutional Neural Network (CNN) was implemented completely from scratch
using TensorFlow and Keras to classify celestial images.

The architecture consists of three convolutional blocks. Each block contains:

• A convolution layer to extract spatial features
• A ReLU activation function to introduce non-linearity
• A max-pooling layer to reduce spatial dimensions

After feature extraction, the network uses:

• A flatten layer to convert feature maps into a 1D vector
• A fully connected dense layer for learning high-level representations
• A final softmax layer to perform multi-class classification

All model weights were initialized randomly and trained from scratch.
No pretrained models (such as ResNet, VGG, or EfficientNet) or external weights
were used.

The model was trained using the Adam optimizer and categorical loss, with
performance evaluated on a validation split of the dataset.
The primary goal of this implementation is to understand the internal working
of CNNs rather than to maximize accuracy.

[OpenGitBot]

Hey @yogesh4216

Thanks for opening this PR 🚀. Mentor will review your pull request soon and till then, keep contributing and stay calm.

Thanks for contributing in OpenCode'25 ✨✨!

[OpenGitBot]

Hey @yogesh4216

Your PR has been merged 🥳🥳 and you have earned 30 points.

Thanks for contributing in OpenCode'25✨✨