Team member: Bingquan Cai, Chuwei Chen, Jiawei Zhao, Xiaowei Ge, Yuxiang Wan
Recognizing the cell identity is usually through slow and costly sequencing-based methods. Image-based methods for cell identification are cost-efficient and fast, but usually difficult to realize by relating the 2D shape information to identity. By applying the machine learning methods, we could extract features more efficiently and improve the image-based methods classification precision.
In recent years, machine learning techniques are widely implemented to solve image classification problems. In this project, we implemented a hybrid model, CNN-SVM, where CNN plays the role of feature extractor which can learn from the data set, and SVM plays the role of a generalized classifier. We showed that our hybrid model improved the classification accuracy compared to each method separately. To benchmark our model performance, we compared the accuracy of the classical MNIST data set. And for application, we demonstrated the advanced accuracy of our model on biological samples to fill the need for precise image-based methods for cell identification.
Sunshineatnoon. (n.d.). Sunshineatnoon/two-layer-CNN-on-mnist. GitHub. Retrieved Apr 24, 2022, from https://github.com/sunshineatnoon/Two-Layer-CNN-on-MNIST.git