Image Classification with the RedEdge-MX Dual Camera System

View sample data hereMore wavebands = More information = More analysis potential.Sometimes, the best way to convey something is to show it visually. This is why were publishing this series of mini-analyses of 10 band data from the RedEdge-MX Dual Camera System.

We flew the Dual Camera over a variety of terrain, from orchards in Arizona to a stream-bed in Washington state, and each dataset came back with an interesting story to tell. Weve detailed them here in hopes that you can see for yourself how 10 band data will power the future of vegetation management.Farmland, Forestry and StreamsBelow is an RGB mosaic from a Dual Camera dataset.

In it, there are a multitude of environments ranging from horticulture to forested areas to water. Even in the RGB imagery, its clear that there are different species of trees, different types of crops, and potentially even some vegetation in the water.The different types of trees and crops are especially visible in this NIR Red Edge Green composite (shown below).

Like we mentioned above, one of the most interesting things about this dataset is the wide variety of features it encompasses. Therefore, the first thing we wanted to do was see if we could easily and accurately classify the different features.Using the full 10 band GeoTIFF, we trained an algorithm to classify this image into 12 different categories: soil, sand, water, grass, trees, dead vegetation, yellow grass, green crop, dark red crop, dark green crop, yellow/bright crop, and potential artifacts.

The results were quite accurate as you can see. The computer was able to classify the entire dataset even though there were some harsh shadows obscuring some features.Were especially excited to see how classification models improve as we add more bands.

With 10 band as opposed to 5 band data there is inherently more information to train models with, opening up the possibilities for classification algorithms and other analysis tools.From here, we would like to take this further and use the different bands to classify the different tree species. This isnt the last time well post about this dataset so stay tuned!

Well be posting mini-stories from other datasets as well, so check back here for updates! In the meantime, you can download RedEdge-MX Dual Camera sample data here, and you can learn more about the system itself here!.