Merge pull request #620 from Karthik110505/main

[Project Addition]: Waste Classification using DL

Waste Classification using DL/Dataset/README.md

@@ -0,0 +1,9 @@
+# Waste Classification Using Deep Learning
+
+The dataset used in this project is taken from the Kaggle website.
+
+**Dataset Link:** [Waste Classification Data](https://www.kaggle.com/datasets/techsash/waste-classification-data)
+
+This dataset consists of images of organic (O) and recyclable (R) waste. The images are labeled to provide a comprehensive dataset for training and evaluating machine learning models.
+
+The task is to classify each waste image into one of the two predefined categories indicating either organic or recyclable waste.

Waste Classification using DL/Model/README.md

@@ -0,0 +1,92 @@
+# Waste Classification Using Deep Learning
+
+**GOAL**
+
+To classify images from the waste classification dataset using a deep learning approach.
+
+**DATASET**
+
+[Waste Classification Data](https://www.kaggle.com/datasets/techsash/waste-classification-data)
+
+**DESCRIPTION**
+
+The dataset contains images of organic (O) and recyclable (R) waste. The task is to classify these images using deep learning architectures.
+
+**WHAT I DID**
+
+First, I imported all the required libraries and the dataset for this project. I split the dataset into training, validation, and testing sets. Then I proceeded to build and evaluate the models.
+
+I developed several deep learning models to classify the images. Initially, I used an Artificial Neural Network (ANN), followed by a Convolutional Neural Network (CNN). Both models did not yield satisfactory accuracy. I then used VGG16, ResNet, and MobileNetV2 architectures, with the latter providing better performance. Finally, I evaluated the performance of all the models in order to pick the optimal model.
+
+**MODELS USED**
+
+The models are:
+
+1. Artificial Neural Network (ANN)
+2. Basic Convolutional Neural Network (CNN)
+3. VGG16 Model
+4. ResNet Model
+5. MobileNetV2 Model
+
+**LIBRARIES NEEDED**
+
+- tensorflow
+- matplotlib
+- opencv-python
+- numpy
+- random
+- shutil
+
+**VISUALIZATION**
+
+### Model 1 (ANN Model) Performance Graphs
+![Model 1 (ANN Model) performance graphs](../Images/ANN_Performance.png)
+
+### Model 2 (Basic CNN Model) Performance Graphs
+![Model 2 (Basic CNN Model) performance graphs](../Images/CNN_Performance.png)
+
+### Model 3 (VGG16 Model) Performance Graphs
+![Model 3 (VGG16 Model) performance graphs](../Images/VGG16_Performance.png)
+
+### Model 4 (ResNet Model) Performance Graphs
+![Model 4 (ResNet Model) performance graphs](../Images/ResNet50_Performance.png)
+
+### Model 5 (MobileNetV2 Model) Performance Graphs
+![Model 5 (MobileNetV2 Model) performance graphs](../Images/MobileNetV2_Performance.png)
+
+**ACCURACIES**
+
+| Model              | Architecture                | Accuracy in % (on testing data) |
+|--------------------|:---------------------------:|:------------------------------:|
+| Model 1            | ANN Model                   | 81.73                          |
+| Model 2            | Basic CNN Model             | 87.31                          |
+| Model 3            | VGG16 Model                 | 90.65                          |
+| Model 4            | ResNet Model                | 76.88                          |
+| Model 5            | MobileNetV2 Model           | 90.85                          |
+
+**CONCLUSION**
+
+After training and evaluating various deep learning models on the waste classification dataset, the results are as follows:
+
+**Key Observations:**
+
+1. **Artificial Neural Network (ANN) Model**: Achieved an accuracy of 81.73%. While it provided a decent baseline, it was outperformed by the other convolutional neural network-based models.
+
+2. **Basic CNN Model**: Improved accuracy significantly to 87.31%, demonstrating the power of convolutional layers in image classification tasks.
+
+3. **VGG16 Model**: Achieved an impressive accuracy of 90.65%. This pre-trained model showed strong performance, indicating the benefits of transfer learning.
+
+4. **ResNet Model**: Obtained an accuracy of 76.88%, which was lower than expected. This might be due to overfitting or insufficient fine-tuning.
+
+5. **MobileNetV2 Model**: Achieved the highest accuracy of 90.85%. This lightweight model is particularly suitable for deployment in resource-constrained environments.
+
+**Conclusion:**
+
+The MobileNetV2 model emerged as the best-performing model for the waste classification task, with an accuracy of 90.85%. Its balance of high accuracy and computational efficiency makes it an excellent choice for practical applications. The VGG16 model also performed very well and could be a suitable alternative depending on the specific use case requirements.
+
+Further improvements can be made by fine-tuning the models, exploring additional data augmentation techniques, and experimenting with different hyperparameters.
+
+**Connect with Me**
+
+- [LinkedIn](https://www.linkedin.com/in/barrenkala-veera-venkata-karthik-b58b9a285/)
+- [GitHub](https://github.com/Karthik110505)