I am trying to classify parking lots and roads with two subcategories each - light and dark. I tried using the support vector machine on ArcGIS with nearly 200 samples of each class, leaving the "Maximum Number of Samples per Class" as the default value 500.
The results decent for the remaining classes (others and Roof (light and dark)). However, the model is getting confused between parking lots and roads due to a significant overlap of features such as similar orientation of cars, similar pixel colors and road markings.
Considering these aspects, what would be the best approach to classify them from a supervised, unsupervised or a deep learning approach?
PS: The classification in total has these classes : Dark Roof, Light Roof, Dark Road, Light Road, Dark Parking, Light Parking and Others (including greenery as it is not a subject of interest)
Use CNN or pre trained model or some trainable state of art model with adam or nestrov sgd optimizer.
One way to classify roads and parking lots using remote sensing data would be to use a machine learning algorithm. This could involve training the algorithm on a large dataset of labeled remote sensing images, where the roads and parking lots have been manually identified. The algorithm could then learn to recognize the characteristics of roads and parking lots in the images, and use this knowledge to make predictions on new, unseen images.
Another method for classifying roads and parking lots in remote sensing images would be to use image processing techniques. This could involve applying various filters and algorithms to the images to extract features such as the shape, size, and texture of the roads and parking lots. These features could then be used to classify the roads and parking lots based on predetermined criteria
- Badges
- Science topic