Plant diseases and pests can significantly impact crop yields and quality, leading to economic
losses for farmers. Early detection and accurate identification of these issues are crucial for
effective management and prevention strategies. Image classification using computer vision techniques
plays a vital role in this process, enabling farmers to quickly identify diseases and pests from
images captured by drones or cameras.
By training custom image classifiers, farmers can leverage machine learning algorithms to recognize
specific diseases and pests based on visual patterns in plant images. These classifiers can be trained
on images of healthy and diseased plants, allowing them to distinguish them based on visual features.
Once trained, these classifiers can analyze new images in real time, providing farmers with instant
feedback on the health of their crops.
How to train a custom image classifier in 5 minutes
DeepLab provides an intuitive interface for managing datasets, annotating images, and training custom
image classifiers, allowing users to create models that can accurately classify images based on specific
criteria. In this tutorial, we will walk you through the process of training a custom image classifier
using DeepLab, focusing on a practical example of classifying screws as either good or bad.
Follow these steps to achieve this:
1. Open DeepLab and create a new folder
Sign in to your Abraia account and go to
DeepLab, to start creating a new folder (related to your
classification task) called "screws." This is where you'll organize your dataset and training-related files.
2. Download the dataset and upload the images
Download the screws dataset and unzip the files to a location on your local machine. Then, click the
"upload" button to select and upload the images from the unzipped dataset to the "screws" folder in DeepLab.
3. Annotate the screws images as good and bad
Navigate to the "screws" folder in DeepLab. Select images that represent good screws clicking on the checkboxes
in every thumbnail. After selecting the images, click on the "annotate" button, and enter "good" as label. Select
the remaining images, that represent bad screws (those not annotated as good). Click on the "annotate" button
again, and enter "bad" as new label.
4. Configure, train, and save the classification model
Click on the "train" button to start the training process. A notebook interface will open. Input your User ID
and the API Key in the notebook. These credential are necessary for accessing the Abraia's training resources.
Inside the notebook, locate the cell that handles the dataset variable. Change the dataset value to "screws".
Click "run all" to execute the entire notebook, which includes loading the dataset, defining the model, and
training.
Once the notebook finished execution, the trained model will be saved in the "screws" folder.
5. Test the model in your own device with DeepLab
Navigate to the "screws" folder and select the images you want to test. Then, double-click on one of the
selected images to open the testing interface. DeepLab will display the model's predictions, indicating whether
the screws are classified as "good" or "bad".
Congratulations! You have successfully trained a classifier for screws using DeepLab. This tutorial provides a
basic overview, and you can further explore advanced features, optimize your model, and fine-tune parameters based
on your specific use case.
