I am working with biologists to automate phenological measurements in the high arctic where data is collected as plant species counts. At the moment the process is mostly manual with very little automation taking place in the form of PDA applications and documenting images. We envision a Computer Assisted Plant Recognition system that uses Wireless Imaging Sensor Networks (WiSN) to automate the identification process. It could potentially provide Biologists with important data such as pheno-phase initialization and finalization dates, individual counts and environmental information (cloud cover, amount of rain, fogg…)
- The species of interest are small and some share their color with the background. This presents a challenge in terms image processing and can have an effect on accuracy.
- Different stages within species can be very similar and difficult to differentiate in an image.
- Harsh conditions in the arctic present a challenge for deployment. The WiSN needs to consider below zero temperatures, high winds and interactions with animals. The location of the plots are scattered in a valley of approximately 4Km² that contains potential interfering elements like small hills.
We have divided the project into 2 subparts:
- The first consists in getting the flower detection to a point that it can be comparable to the level of accuracy that the biologists have at the moment. We will also strive to increment the level of accuracy and reliability of other automated flower detectors. This phase does not include any WSN component. We are working with images taken from the biologists’ cameras which we then feed into a regular PC via storage cards. We are currently experimenting with state of the art computer vision methods to detect the flowers.
- The second phase (Still in planning) consists in brining what we are doing in the first phase into a WiSN. We expect to have to deal with high computational problems, as the computer vision algorithms are CPU hungry. We also expect to centre this second phase in WSN reliability which we think will be a relevant issue given the harsh Arctic conditions.