Merge pull request #547 from CoderOMaster/main

Cassava Leaf Disease Classification
abhisheks008 · Jan 25, 2024 · ed4e2db · ed4e2db
2 parents 3f3cb7f + 52fbc15
commit ed4e2db
Show file tree

Hide file tree

Showing 10 changed files with 1,205 additions and 0 deletions.
diff --git a/Cassava Leaf Disease Classification/Dataset/README.md b/Cassava Leaf Disease Classification/Dataset/README.md
@@ -0,0 +1,3 @@
+https://drive.google.com/drive/folders/1KLfnDtRlBaVWG_5xMfst-d9M9GhYTLno
+
+Uploaded due to GITHUB limit.
diff --git a/Cassava Leaf Disease Classification/Images/1.png b/Cassava Leaf Disease Classification/Images/1.png
diff --git a/Cassava Leaf Disease Classification/Images/2.png b/Cassava Leaf Disease Classification/Images/2.png
diff --git a/Cassava Leaf Disease Classification/Images/Example.png b/Cassava Leaf Disease Classification/Images/Example.png
diff --git a/Cassava Leaf Disease Classification/Images/Plot.png b/Cassava Leaf Disease Classification/Images/Plot.png
diff --git a/Cassava Leaf Disease Classification/Model/CNN (1).ipynb b/Cassava Leaf Disease Classification/Model/CNN (1).ipynb
diff --git a/Cassava Leaf Disease Classification/Model/Misc_Models.ipynb b/Cassava Leaf Disease Classification/Model/Misc_Models.ipynb
diff --git a/Cassava Leaf Disease Classification/Model/UnderSampled.ipynb b/Cassava Leaf Disease Classification/Model/UnderSampled.ipynb
diff --git a/Cassava Leaf Disease Classification/README.md b/Cassava Leaf Disease Classification/README.md
@@ -0,0 +1,69 @@
+# CASSAVA LEAF DISEASE CLASSIFICATION
+
+## GOAL
+Developing various computer vision models to classify leaf diseases.
+
+## DATASET
+https://www.kaggle.com/datasets/nirmalsankalana/cassava-leaf-disease-classification
+
+## MODELS USED
+- CNN
+- VGG16
+- Inception
+- ResNet50
+- AlexNet
+
+## LIBRARIES
+- Pandas
+- Numpy
+- TensorFlow
+- OS,Shutil
+- Matplotlib
+- Scikit-Learn
+- Seaborn
+
+## IMPLEMENTATION
+1. Load dataset (21,000 entries and 5 columns)
+2. Implemented Deep learning models.
+3. Applied data augmentation and undersampling of dataset separately for comparison.
+4. Alexnet and undersampled models werent trained fully due to computational and time constraints but their expected accuracy is mapped logically.
+
+
+## Models and Accuracies
+
+| Model             | Accuracy   | Validation Loss |
+| ----------------- |:----------:|:---------------:|
+| CNN               | 0.70       |     0.808       |
+| VGG16             | 0.61       |     0.911       |
+| CNN(Data Aug)     | 0.66       |     0.865       |
+| ResNet            | 0.61       |     1.099       |
+| InceptionV3       | 0.69       |     0.833       |
+| AlexNet           | ~0.62      |     ~1.000      |
+| CNN(Undersampled) | ~0.5       |     ~1.00       |
+             
+
+**VISUALISATION**
+
+![Alt Text](./Images/1.png)
+
+EDA Analysis as Bar Graph(Before Undersampling)
+
+![Alt Text](./Images/2.png)
+
+EDA Analysis as Pie Chart(Before Undersampling)
+
+![Alt Text](./Images/Plot.png)
+
+After Effective Undersampling from 17k to 2k ish in last class of mosaic_disease.
+
+![Alt Text](./Images/Example.png)
+
+Example of Mismatch for demonstration purpose.
+
+**CONCLUSION**
+
+CNN Model is the best out of all models(accuracy and validation loss).In this database undersampling and data augmentation were found to not have any significant impact.
+
+**NAME**
+
+Keshav Arora
diff --git a/Cassava Leaf Disease Classification/requirements.txt b/Cassava Leaf Disease Classification/requirements.txt
@@ -0,0 +1,6 @@
+pandas==1.3.3
+matplotlib==3.4.3
+numpy==1.21.2
+tensorflow==2.6.0
+torch==1.9.1
+transformers==4.10.3