Segment Aerial Photos with an AI Model Trained on Drone Photos

In this article, we will have a machine learning model trained on small, low-resolution drone photos segment larger aerial photos.
Deep Block is a powerful AI model development tool that strengthen machine learning models trained on small drone images to analyze large aerial and satellite images.

Everything Starts from Small Things

From humble beginnings, greatness emerges. All we had initially were low-resolution drone photos.
With this data, we created a machine learning model in Deep Block.
The images we used to train the model were small drone photos.

In our previous endeavor, we successfully built a machine learning model capable of analyzing high-resolution 4K drone images using these smaller images as training data.

However, WE ARE OMNIS LABS COMPANY. We have the attitude to EXPERIMENT(Labs) EVERY(OMNIS)THING.

Our curiosity began with the following question.
“How about analyzing LARGE aerial photos with our machine learning model trained on small drone photos?”

So, we started the EXPERIMENT.

Project preparation

Let’s prepare the trained AI project we created last time.

Copy the project and open the PREDICT tab.

And we first deleted the existing images in the predict tab.

Uploading New Images

To assess the capabilities of Deep Block, we gathered LARGE aerial photographs for experimentation. These images boast a size of around 79MB and a resolution of approximately 4600x5600 pixels.

The test image can be downloaded here.

Drag & Drop the image to the file explorer.

Divide Images

Let's divide our large aerial image.

• Click Configuration Icon.
  
  
  
• Next, we proceed to split the image into a grid of 16 rows and 15 columns.
  
• Then click DIVIDE button.

Let's Detect

• Click PREDICT button to start the inference.
  
  
  
• It will take some time to accurately segment a high-resolution aerial image, so be patient.
  If you want to make inferences faster, contact us.
• Something happened.
  
• We succeeded in segmenting large aerial photos with a machine learning model trained on small drone photos.
  

What's so great about this?

We created a machine learning model that analyzes LARGE aerial and satellite images without aerial or satellite images.

Acquiring high-resolution aerial or satellite images can be a costly and challenging endeavor, with numerous legal regulations.

Usually the government has these, and these images are either inaccessible or expensive in the market.

However, we can take images of the ground with a small drone.
Moreover, drone photographs and videos are abundantly available across the internet.

We managed to create a machine learning model that analyzes BIG images using a machine learning model trained on images with less than 1K resolution.

The training images we used are not expensive and easy to find online.
We can create computer vision for remote sensing images without purchasing large aerial and satellite images.

And, Deep Block made it possible.
Some people ignore small things such as small company, small photos.
However, the power of small things should never be underestimated as they have the potential to achieve greatness.

As you have witnessed the incredible capabilities of machine learning models trained on small images, it is evident that Deep Block has much more to offer.

Stay tuned for my upcoming article where we will show the other remarkable powers that Deep Block possesses.