Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/mk2345/gtsrb-dl-ml

Image Classification using CNN and SVM (Keras & Scikit-Learn)
https://github.com/mk2345/gtsrb-dl-ml

jupyter-notebook keras-tensorflow scikitlearn-machine-learning

Last synced: 5 days ago
JSON representation

Image Classification using CNN and SVM (Keras & Scikit-Learn)

Awesome Lists containing this project

README 

        
## Traffic Sign Classification using Deep Learning and Machine Learning



This project uses the German Traffic Sign Recognition Benchmark (GTSRB), a large dataset of more than

50,000 traffic sign images in 43 classes. The GTSRB is widely used for evaluating machine learning

and computer vision algorithms in the context of traffic sign recognition.



### Model Development and Optimization



Multiple SVM and CNN models were trained with varying amounts of training data to determine the threshold

at which one model type outperforms the other. Additionally, to ensure optimal performance, Grid Search was

utilized for hyperparameter tuning of select models.

Depending on the specific model requirements, image preprocessing techniques were applied, including image

enhancement and feature extraction. These preprocessing steps were implemented to improve the quality of input 

data and to extract relevant features, thereby potentially enhancing the overall performance of the models.



### Below are some sample images of the dataset.



![gtsrbDatensatz](https://github.com/MK2345/GTSRB-DL-ML/assets/24621381/bcfce9d8-655d-4837-8be3-2791e8775f92)



### Class distribution:



The training dataset is imbalanced, with significant variations in the number of samples across different

traffic sign classes. This imbalance is an important characteristic of the GTSRB dataset and reflects the

real-world frequency of different traffic signs.



Key points about the class distribution:

1. Some classes are overrepresented with thousands of samples.

2. Other classes are underrepresented with only a few hundred samples.

3. This imbalance can affect model performance and requires careful consideration during training and evaluation.



To address this imbalance, I employed techniques such as using class weights and Data augmentation.



![stratified_distribution_39209_gtsrb](https://github.com/MK2345/GTSRB-DL-ML/assets/24621381/47592cfc-13ac-4c51-a170-40288825d2e6)



### Stratified Class Density Visualization (500 Images)

![plot_500](https://github.com/MK2345/GTSRB-DL-ML/assets/24621381/b4c90437-8c86-4cca-9812-7cb90bfb6acd)



### Stratified Class Density Visualization (whole dataset)

![plot_39209](https://github.com/MK2345/GTSRB-DL-ML/assets/24621381/0df62558-6641-41bc-8b1a-6f9e64779174)



### Histogram equalization



The pictures were taken under various weather conditions, and the contrast of low-contrast images was enhanced through histogram 

equalization, a technique that redistributes pixel intensity values.



![GTSRB_Hist_eq](https://github.com/MK2345/GTSRB-DL-ML/assets/24621381/02700502-6f0c-4593-b31d-495c28f9d659)

![GTSRB_Hist_eq_v](https://github.com/MK2345/GTSRB-DL-ML/assets/24621381/5611f024-3e9b-4804-a4c7-5163580a99f2)



### HOG-Features



The SVMs were trained using HOG features extracted from the images, among other features



![GTSRB_HOG](https://github.com/MK2345/GTSRB-DL-ML/assets/24621381/19117e85-b5ee-4941-8c28-fb00ee86d7d1)



### Classification results as a function of the training set



The best results were achieved with a CNN with 6 layers and data augmentation up to the original total amount.








#### Example: minority class 0

On the left the image to be classified, on the right the misclassified image.






#### Classification results of the minority class 0