Deep Learning for Leopard Individual Identification: An Adaptive Angular Margin Approach

TL;DR

A deep learning framework to distinguish between individual leopards based on their unique spot patterns is introduced, employing a novel adaptive angular margin method in the form of a modified CosFace architecture and a preprocessing pipeline that combines RGB channels with an edge detection channel to underscore the critical features learned by the model.

Abstract

Accurate identification of individual leopards across camera trap images is critical for population monitoring and ecological studies. This paper introduces a deep learning framework to distinguish between individual leopards based on their unique spot patterns. This approach employs a novel adaptive angular margin method in the form of a modified CosFace architecture. In addition, I propose a preprocessing pipeline that combines RGB channels with an edge detection channel to underscore the critical features learned by the model. This approach significantly outperforms the Triplet Network baseline, achieving a Dynamic Top-5 Average Precision of 0.8814 and a Top-5 Rank Match Detection of 0.9533, demonstrating its potential for open-set learning in wildlife identification. While not surpassing the performance of the SIFT-based Hotspotter algorithm, this method represents a substantial advancement in applying deep learning to patterned wildlife identification. This research contributes to the field of computer vision and provides a valuable tool for biologists aiming to study and protect leopard populations. It also serves as a stepping stone for applying the power of deep learning in Capture-Recapture studies for other patterned species.

Figures (14)

• Figure 1: (a) Four images of a leopard in different conditions. (b) Distribution of images per leopard flank.
• Figure 2: Full Preprocessing pipeline. Extract bounding box, remove background, edge detection. Input cropped RGB and edge detection to the model.
• Figure 3: Triplet Network in training phase with Semi-hard negative mining: Negative embedding selected must be in the red area to satisfy condition. In the testing/inference phase, embeddings produced by the network are directly compared.
• Figure 4: CosFace model representation. In the training phase, an extra fully-connected layer is added. The layer is discarded in testing, where embeddings are directly compared in the hypersphere through cosine similarity.
• Figure 5: Showing how CosFace ensures class separation on a random selection of chosen leopards.