ResNet-50, EfficientNet, and Siamese Neural Networks are used in unison to identify tigers through biological markers instead of mere raw pixels. To understand how this works, read below: Image Capture A trap camera photographs a tiger walking on a trail. The focus is to capture the tiger from different angles. Feature Extraction The features of the tiger are extracted using ResNet-50 or EfficientNet, which does not memorize the entire photograph. Rather, it focuses on unique markers such as stripe spacing, stripe patterns on the torso, facial markings, tail rings, and ear notch shapes. These are then classified as vector features using a numerical form: Tiger Image → ResNet-50 / EfficientNet → Feature Vector [0.34, 0.81, 0.19, 0.65, ...] Comparison Using Siamese Neural Network When the image of another tiger is captured, it is compared with the existing database. The feature vectors are compared and comparison is made, which generates a similarity score. Image A → ResNet/EfficientNet → Feature A \ ► Similarity Score / Image B → ResNet/EfficientNet → Feature B The model then compares the two feature vectors. Similarity = 0.98 → Same tiger Similarity = 0.42 → Different tiger The tiger is then named based on its unique features. To put it in easier context, the tiger identification workflow occurs in this way: Trap cameras are used to collect tiger images in motion. ResNet-50 or EfficientNet models are used to convert image into feature vectors. A Siamese Neural Netowrk compares feature vectors against the database of identified tigers. If a close match is found, the tiger is identified from the database. If no match exceeds the similarity threshold, the system flags it as a potentially new individual, which is then verified with the similar exercise. Tiger identification, assisted with new-age technology, gives accurate results with no to least disruptions to animal movement and help maintain a database required for conservation efforts. Share this article