Jewellery Classification

Introduction

Jewellery Classification is a Convolutional neural network (CNN ) model that designed for 5 different jewellries such as Bracelet, Earrings, Necklace, Rings, Wristwatch and able to predict sample test data.

Dataset

sn	Name	Training	Test
1	Bracelet	355	50
2	Earrings	676	50
3	Necklace	251	50
4	Rings	183	50
5	Wristwatch	121	50

Architecture

1. Convolutiona 2D layer with a specific requiring parameters such as filter:32, kernel_size: (3,3) image size
2. MaxPooling to reduce number of features with pool size (2,2).
3. Flatten layer to flatten matrix to vector so that it can be used in a dense layer
4. Application of a dropout of 0.2% to avoid over fitting with an activation function of relu.
5. Dense layer/Hidden Layer or a fully connected layer with a neurons of 128
6. Added a dropout of 0.2 to keep the model from over fitting
7. Hidden layer of 128 neurons with an activation function of relu.
8. Added a dropout of 0.2 to keep the model from over fitting.
9. finally added an output layer with a unit of 5 neurons(number of classes of datasets) and a softmax activation function.

Image Augmentation

Due to the size of our dataset and class imbalance, we will not get the right accuracy. there is a need to increase the size of our dataset to get optimal result.

Image Augmentations techniques are methods of artificially increasing the variations of images in our data-set by using horizontal/vertical flips, rotations, variations in brightness of images, horizontal/vertical shifts etc.

Keras ImageDataGenerator class is used to perform this operation.

Hyperparameter

• batch_size = 25
• epoch = 25
• steps_per_epoch = 15
• optimizer = adam
• loss = categorical_crossentropy
• metrics = ['accuracy']

Result

img size	Training (loss)	Training(acc)	Test (loss)	Test(acc)
32px vs 32px	0.2955	0.9013	0.6947	0.8160
64px vs 64px	0.2733	0.9040	0.4626	0.8560
128px vs 128px	0.2389	0.9333	0.5702	0.8507

Install Dependencies

1. Clone the repository on your system
2. Install the necsessary packages such as
   • Python2 or Python3
   • Tensorflow
   • Keras
   • Numpy

Run Program

python classifier.py