Scientists have come up with a new automated method to prepare digital photos of animals to help in the analysis done by wildlife researchers. Frequently, these researchers depend on photographs in identifying individual animals by their unique markings. Together with scientists from Microsoft Azure, wildlife biologists from Penn state they improved how photos are turned to usable data through a cloud computing service and machine learning
Derek Lee, an associate research professor of biology, said that researchers require to identify and collect data for their work hence instead of human-applied markings and tags which could interfere with the animal’s behavior, these researchers take photographs of the animals. As much as there are software used to analyze the photo, they, however, need to be manually prepared for analysis which is time-consuming.
Lee uses photographs to understand births, deaths, and movement of more than 3,000 giraffes in East Africa. He collaborated with the scientists from Microsoft who provided him with a new image processing machine services. The service was essential for automating the time-consuming process that Lee and his team had to go through, by using machine learning technology on the Microsoft Azure cloud.
By using a computer algorithm for objective detection, the team trained a program to recognize some of the giraffe’s torsos through using the existing photos. The program improved using an active learning process whereby the system was able to show predicted cropping squares on new images to a human who could immediately verify or correct results.
The new images were fed back into the training algorithm for further improvement and updating of the programme. The system could identify the location of giraffe torsos win the picture with higher accuracy even if the giraffe is a small portion of the photo or its torso is semi-blocked by vegetation.