Objectives for Canopify

Our client, Southern Cross Drones (SCD), provides a broad range of drone services for aerial imagery. In our case aerial images of a vineyard and macadamia nut farm have been captured showing vegetation canopy. Analysis of images is required to identify areas of poor canopy health as an indicator of plants or areas requiring special attention and repair.

Canopify is an algorithm developed for SCD. It is intended for deployment on remote server access through an established platform similar to DroneDeploy, an affiliate of SCD. 

Some initial work was undertaken in MatLab and provided to the group.

What’s the idea?

As the algorithm is required to work on a remote server, previously developed work is to be ported to Python as it can be run on the remote server through Flask.

To make the assessment a process of decorrelating, binarizing, sectioning and brute force counting relevant pixels then ranking those counts via a heat map indicating good and poor canopy health.

• Decorrelating the images assists with separating colours of canopy from the background.
• Binarising exacerbates the contrast of pixels included in the counting.
• Sectioning establishes the rows of plants and area within the row to provide the feedback of the assessment (Hear Map).

Furthermore, morphological processes were applied to define pixel clusters with OpenCV and numpy. 

Machine learning algorithms were implemented to build a model that would directly show which plants were unhealthy. ImageAI, LabelME and LabelImg were used for preprocessing and development of the image processing model.

The Team

I have been fortunate to work with an amazing team of academics and fellow students from the School of Data and Electrical Engineering (SEDE) and the Faculty of Engineering and Information Technology (FEIT). The team comprised of the following members: