Algorithm for jaguar detection from camera traps
Data collected by camera traps in the form of images were used to detect jaguars. The data was partitioned into 3 sets: training, validation and testing. Accuracy, recall, precision and F1 metrics were calculated. The most relevant metric is F1 which describes how effective the algorithm is in detecting a jaguar in the image (Test accuracy : 0.906, Test recall : 0.863, Test precision : 0.899, F1 : 0.87). Likewise, 4 databases were created within the Huawei Cloud (Jaguar presence during the day, presence of jaguar at night, absence of jaguar during the day, absence of jaguar at night). Finally, 3 algorithms were tested to compare which one effectively detected the presence and absence of jaguar by day or night.
Classifications
Category
Scale of implementation
Phase of solution
Enabling factors
Data collected from the devices and strategic alliances with Huawei and the Polytechnic University of Yucatan.
Lessons learned
The algorithm has over 90% accuracy and is trained on photos from camera traps. The identified risks include potential failure on photos that are outside the training spectrum (blurry, too dark, too light or images where the jaguar's body is obstructed by some other object). Mitigation strategies:
-
Training the model also with other types of images other than camera traps or with more images to broaden the spectrum.
-
Database and algorithm feedback from peer-reviewers and expert collaborators with relevant experience in monitoring felines and their prey will be added as peer-reviewers, providing jaguar images and expert feedback.
Part of the panorama areas de conservación solution
