Contour-Based Wild Animal Instance Segmentation Using a Few-Shot Detector

Abstract

Camera traps are widely used in wildlife research, conservation, and management, and abundant images are acquired every day. Efficient real-time instance segmentation networks can help ecologists label and study wild animals. However, existing deep convolutional neural networks require a large number of annotations and labels, which makes them unsuitable for small datasets. In this paper, we propose a two-stage method for the instance segmentation of wildlife, including object detection and contour approximation. In the object detection stage, we use FSOD (few-shot object detection) to recognize animal species and detect the initial bounding boxes of animals. In the case of a small wildlife dataset, this method may improve the generalization ability of the wild animal species recognition and even identify new species that only have a small number of training samples. In the second stage, deep snake is used as the contour approximation model for the instance segmentation of wild mammals. The initial bounding boxes generated in the first stage are input to deep snake to approximate the contours of the animal bodies. The model fuses the advantages of detecting new species and real-time instance segmentation. The experimental results show that the proposed method is more suitable for wild animal instance segmentation, in comparison with pixel-wise segmentation methods. In particular, the proposed method shows a better performance when facing challenging images.

MammalClub: An Annotated Wild Mammal Dataset for Species Recognition, Individual Identification, and Behavior Recognition

“…In [54], we adopted ResNet and RPN for feature extraction and region proposals, respectively. We allocated the training and evaluation set at a ratio of 0.85, and the details are shown in [54].…”

“…In [54], we adopted ResNet and RPN for feature extraction and region proposals, respectively. We allocated the training and evaluation set at a ratio of 0.85, and the details are shown in [54]. In the training stage, we introduced the few-shot training strategy in a convolutional neural network for FSOD (few-shot object detection) in order to recognize new mammal species that had a small number of samples in the training set.…”

“…We proposed novel methods based on deep learning for species recognition, individual recognition, and action recognition, respectively [54,60,61]. We conducted experiments on our dataset with the proposed methods and other state-of-the-art methods for a comparison under the same experimental configuration.…”

“…High similarity between consecutive frames leads to data redundancy; thus, the study used the histogram method to calculate the similarity between images and eliminated similar images in the same video clip [54]. The gray-scale histogram counts the number of pixels at a certain grayscale level, as seen in Figure 2.…”

Wildlife Real-Time Detection in Complex Forest Scenes Based on YOLOv5s Deep Learning Network

“…However, it achieves greater accuracy in the task of instance segmentation, resulting in more precise target detection. Jiaxi Tang et al proposed a two-stage model based on the Mask R-CNN model for detecting wildlife targets captured by trap cameras [20]. In the first stage, fewshot object detection is used to identify the species and initially describe the target contour.…”