Page 170 - Emerging Trends and Innovations in Web-Based Applications and Technologies
P. 170
International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
Shape Analysis: Shape-based features can help determine product quality and originality.
Histograms of Intensity: The intensity of colors and pixels can reveal information about a product’s condition.
Spatial Filters: Spatial filters are used to enhance the image quality and highlight important details about the product.
Wavelet Transforms: Wavelet-based features provide a multi-scale analysis of images to capture fine details.
Classification
The classification of second-hand products is carried out using a Convolutional Neural Network (CNN). This deep learning
technique is highly accurate when dealing with image datasets, and it is used to classify products into their respective
categories and predict their price ranges.
The CNN model is trained to classify products based on various features such as condition, type, and brand. The classification
results assist sellers in determining fair prices and help buyers make informed decisions based on product conditions and
descriptions.
IV. PROPOSED RESEARCH MODEL
The proposed work leverages a convolutional neural network (CNN) model to enhance Quick Mart's approach in
revolutionizing the second-hand marketplace. CNN, a powerful deep learning structure, is widely used for classification tasks
such as image recognition and object detection. In the context of Quick Mart, this model could be employed to categorize
various second-hand products efficiently. The model will be designed to classify products into categories based on images,
enabling a seamless user experience for both buyers and sellers.
Quick Mart's platform requires efficient and accurate product categorization to enhance the overall user experience, and a CNN
model can help in achieving this. By utilizing this structure, Quick Mart can automatically classify second-hand products into
distinct categories, such as electronics, furniture, fashion, books, etc., making it easier for users to navigate the marketplace.
The architecture of the model involves several layers that process the input images and produce output in the form of class
probabilities. The layers are sequentially organized, where the output of one layer is used as input for the next layer.
1. Conv2D Layer: The first layer in the CNN model is the Conv2D layer, which performs the convolution operation on the
input image using a set of learnable filters. For this case, 32 filters of size 3x3 are used, with 'relu' (rectified linear unit)
activation, which is a common choice in CNNs for efficient learning.
2. MaxPooling2D Layer: The next layer is the MaxPooling2D layer, which performs down-sampling by selecting the
maximum value from a window of size 2x2 in the input image. This layer helps reduce the spatial dimensions of the output,
allowing the model to focus on the most important features.
3. Repeat Layers: The Conv2D and MaxPooling2D layers are repeated with an increased number of filters (e.g., 64 filters),
maintaining the same kernel size and activation function.
4. Flatten Layer: After the convolutional layers, a flatten layer is employed to transform the multi-dimensional output into a
one-dimensional array. This transformation makes the data suitable for further processing by fully connected layers and
ensures that the extracted features can be efficiently classified.
5. Dense Layer: The next step is the dense layer, which is a fully connected layer with a 'relu' activation function. This layer is
responsible for processing the features extracted by the convolutional layers and helps make predictions.
6. Final Softmax Layer: The final layer uses the 'softmax' activation function to output class probabilities for each category,
allowing the model to determine which category a given product belongs to.
The model is compiled using the 'categorical_crossentropy' loss function, the 'adam' optimizer, and accuracy as the evaluation
metric. It is trained for 10 epochs with a batch size of 32, and the training dataset is split into training and validation sets with
an 80:20 ratio. After training, the model is evaluated on the test set, and the test loss and accuracy are reported. The model is
then saved for future use.
In this context, the CNN architecture enables Quick Mart to effectively categorize second-hand products, with the model
achieving an accuracy of 92.14% on the test set, indicating its robustness and efficiency in classifying products within the
marketplace.
V. PERFORMANCE EVALUATION
To evaluate the performance of Quick Mart’s CNN model, a confusion matrix and classification report are used to measure key
metrics such as accuracy, precision, recall, and the F1 score. These metrics help in understanding how well the model is
classifying second-hand products.
Accuracy: This metric measures the proportion of correctly classified instances out of all instances. It is calculated as:
Accuracy= True Positives + True Negatives
Total Items
where TP represents true positives, TN represents true negatives, FP represents false positives, and FN represents false
negatives.
Precision: Precision indicates how often the classifier correctly identifies positive instances. It is calculated as:
IJTSRD | Special Issue on Emerging Trends and Innovations in Web-Based Applications and Technologies Page 160