Page 184 - Emerging Trends and Innovations in Web-Based Applications and Technologies

International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
 Categorical Conversion: Labels are further converted filters. In this case, 32 filters of size 3x3 are used, with a
into categorical format to improve model training rectified linear unit (ReLU) activation function, which is
performance. widely adopted for deep learning due to its effectiveness in
promoting efficient learning.
Feature Extraction
Feature extraction plays a pivotal role in improving model 2. MaxPooling2D Layer:
accuracy. In Quick Mart's case, feature extraction involves After the convolution operation, a MaxPooling2D layer is
identifying and isolating meaningful patterns or employed to perform down-sampling. This layer selects the
characteristics from product images. These extracted maximum value from a window of size 2x2 in the input,
features are essential for accurate product categorization reducing the spatial dimensions of the output and helping
and price prediction. Methods used for feature extraction the model focus on the most relevant features.
include:
3. Repeated Layers:
 Texture Analysis: Features such as entropy and
The Conv2D and MaxPooling2D layers are repeated with an
homogeneity are analyzed to assess product condition
increased number of filters (e.g., 64 filters), maintaining the
and detect any defects.
same kernel size and activation function, allowing the model
 Shape Analysis: Shape-based features help determine to learn increasingly complex features at different levels of
the quality and authenticity of the product. abstraction.
 Histograms of Intensity: Color and pixel intensity 4. Flatten Layer:
levels provide insights into the product's condition.
After the convolutional layers, a flatten layer is used to
 Spatial Filters: Applied to enhance image clarity and transform the multi-dimensional output into a one-
highlight key product details. dimensional array. This transformation makes the data
compatible with the fully connected layers and prepares it
 Wavelet Transforms: These features allow for multi-
scale analysis of images, capturing fine-grained details. for classification.
5. Dense Layer:
Classification
The dense layer is fully connected and utilizes the ReLU
Product classification is performed using a Convolutional
activation function. This layer is responsible for processing
Neural Network (CNN), a deep learning model that excels in
the features extracted from the convolutional layers,
handling image datasets. The CNN is trained to classify
combining them, and making predictions regarding product
second-hand products into appropriate categories and
categories.
predict their price ranges based on features such as
condition, type, and brand. This classification system aids 6. Final Softmax Layer:
sellers in setting competitive prices and assists buyers in The final layer of the model is the Softmax layer, which
making well-informed purchasing decisions based on the outputs class probabilities for each category. This layer
product's condition and description. enables the model to classify products into specific
IV. PROPOSED RESEARCH MODEL categories with high confidence.
The proposed research leverages a convolutional neural Model Training and Evaluation
network (CNN) model to enhance Quick Mart's innovative The model is compiled using the categorical cross-entropy
approach to revolutionizing the second-hand marketplace. loss function, the Adam optimizer, and accuracy as the
CNNs are powerful deep learning structures that are evaluation metric. It is trained for 10 epochs with a batch
extensively used in classification tasks, such as image size of 32. The training dataset is split into 80% for training
recognition and object detection. In the context of Quick and 20% for validation. Once training is complete, the model
Mart, the CNN will be employed to automate the is evaluated on a test set, and its performance in terms of
categorization of various second-hand products, loss and accuracy is reported. The trained model is saved for
streamlining the process for users and enabling a seamless future use, ensuring that Quick Mart can deploy it for
user experience for both buyers and sellers. automatic product categorization.
A critical aspect of Quick Mart’s platform is efficient and Based on the test set results, the model achieves an accuracy
accurate product categorization. The CNN model will address of 92.14%, which demonstrates its robustness and
this challenge by classifying second-hand products into reliability in categorizing second-hand products effectively.
distinct categories, such as electronics, furniture, fashion, V.
and books, based on product images. This automatic PERFORMANCE EVALUATION
categorization system will not only make navigation through To assess the effectiveness of Quick Mart's CNN model,
the platform more intuitive but also improve the overall several key performance metrics are used, including
accuracy, precision, recall, and the F1 score. These metrics
experience for users, helping them find the products they
provide a comprehensive understanding of how well the
need with minimal effort.
model performs in classifying second-hand products.
CNN Model Architecture
The architecture of the proposed CNN model is designed to 1. Accuracy:
process input images and produce output in the form of class Accuracy measures the proportion of correctly classified
probabilities. The CNN consists of several sequential layers, instances out of all instances. It is calculated as:
each responsible for transforming the data and learning Accuracy = True Positives +True Negative
different features of the images. Total Items
1. Conv2D Layer: Precision:
The first layer of the CNN is a Conv2D layer that performs Precision indicates how frequently the classifier correctly
convolution on the input images using a set of learnable identifies positive instances. It is calculated as:

IJTSRD | Special Issue on Emerging Trends and Innovations in Web-Based Applications and Technologies Page 174