Labeling Photos Manually and Classification of Large Dataset

Hello;

Our goal is to download some photos from WEB and classift the musillage part of each photo. Then train a NN for supervised learning to classify any photo that is uploaded. We classified 67 photos, some of them has mussillage, by using image labelling. For training network, and other parts, we couldn't find reliable source(probably our mistake), would you please help how to proceed?

We have an output in the workspace "1x1 groundtruth, a project via Imageclassification" but not sure how to proceed.

Examples of photos are attached.

Thanks a lot in advance.

Answer 1

[Aravind]

Hi @Fatih,

It sounds like you are looking to train a neural network to classify images with and without mussilage. You have already manually labeled 67 images and have a 1x1 groundtruth object in your workspace, likely created using MATLAB’s Image Labeler or a similar tool.

To proceed with creating a neural network for image classification, you can follow these steps:

1. Organize and Prepare Data: Ensure your images are properly labeled. Use the “imageDatastore” function to manage your dataset and split it into training, validation, and test sets. You can find more details in the documentation: https://www.mathworks.com/help/matlab/ref/matlab.io.datastore.imagedatastore.html.
2. Set Up the Neural Network: Consider using a pre-trained network like AlexNet, GoogLeNet, or ResNet for transfer learning. Modify the final layers to fit your classification needs and set training options using “trainingOptions”. Alternatively, you can design your own neural network for classification.
3. Train the Network: Use the “trainnet” function with your training data and specified layers. Monitor the training process and adjust parameters as needed. More information about “trainnet” is available here: https://www.mathworks.com/help/deeplearning/ref/trainnet.html.
4. Evaluate and Fine-Tune: Test the model with your test set and analyze performance metrics such as accuracy and precision. Fine-tune the model by adjusting training parameters to achieve optimal performance.

Here are some resources and tutorials to help you with image classification in MATLAB:

1. Workflow for creating training data: https://www.mathworks.com/help/vision/ug/training-data-for-object-detection-and-semantic-segmentation.html
2. Example of image classification using MATLAB: https://www.mathworks.com/help/vision/ug/image-category-classification-using-deep-learning.html
3. Creating a simple image classifier network: https://www.mathworks.com/help/deeplearning/gs/create-simple-deep-learning-classification-network.html
4. Getting started guide for image classification: https://www.mathworks.com/help/deeplearning/gs/create-simple-image-classification-network-using-deep-network-designer.html
5. Video tutorial for image classification: https://www.mathworks.com/videos/image-classification-with-deep-learning-1701240556615.html
6. Tutorial for creating the datastore for image classification: https://www.mathworks.com/help/deeplearning/ug/create-and-explore-datastore-for-image-classification.html

I hope this helps you get started!

[Fatih]

Great thanks. Actually after some time, thanks to help in Matlab, we successfully classified and tested even with other photos. But a question of possible improvement arose. Is there a way to receive the rate of musillage (I mean in the photo %20 percent, %50 percent musillage exist). The model we made, only confirms or denies the musillage in the photos.