Animal Recognition and Identification with Deep Convolutional Neural Networks for Automated Wildlife Monitoring

Nguyen, Hung Son; Maclagan, Sarah; Nguyen, Tu Dinh; Nguyen, Thin; Flemons, Paul; Andrews, Kylie; Ritchie, Euan G.; Phung, Dinh

doi:10.1109/dsaa.2017.31

Cited by 142 publications

(92 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Our use of transfer learning is a critical departure from other wildlife recognition frameworks which have trained all the weights using the target datasets (Nguyen et al, 2017;Norouzzadeh et al, 2018;Villa TA B L E 5 The performance of the CNN-1 for multiclassification with the resampling process accuracy of CNN-2 for the categories containing more data (badger, bird, and fox) was greater without resampling than with resampling, while accuracy for categories with less data (cat, rat and rabbit) was greater with resampling. An additional advantage was that, since the weights for CNN-2 were already pretrained, the training time of 2,289 s was considerably less than the 6,076 s required for CNN-1.…”

Section: Discussionmentioning

confidence: 99%

“…In this transfer learning process, the already trained weights are used as the initial weights and are then fine-tuned using the task dataset. The assumption is that the network has already learned useful features and could therefore attain greater accuracy than a model trained on a smaller dataset (Nguyen et al, 2017).…”

Section: Deep Learning For Wildlife Species Recognitionmentioning

confidence: 99%

“…The use of remote still and video surveillance cameras in wildlife research and management has grown rapidly in recent years (Nguyen et al, 2017;Villa, Salazar, & Vargas, 2017;Zeppelzauer, 2013). The purpose of surveillance may vary widely from identification of pest species or problem behavior to estimating the abundance and distribution of species of conservation importance, but they usually share a common need, which is to identify particular target species.…”

Section: Introductionmentioning

confidence: 99%

“…For example, Gomez, Salazar, and Vargas (2016) developed a CNN to identify wild animals from the world's largest camera trap project published to date, known as the Snapshot Serengeti dataset (3.2 million images of 48 species; Swanson et al, 2015). Overall accuracy for animal identification was not reported, but was estimated at approximately 57% elsewhere (Norouzzadeh et al, 2018). Very recently, Norouzzadeh et al (2018) applied different CNN architectures including AlexNet (Krizhevsky et al, 2012), VGG (Simonyan & Zisserman, 2015), and ResNet (He, Zhang, Ren, & Sun, 2016) to the same dataset and achieved an accuracy of 92% for species identification.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Wildlife surveillance using deep learning methods

Chen

Little

Mihaylova

et al. 2019

Ecology and Evolution

View full text Add to dashboard Cite

Wildlife conservation and the management of human–wildlife conflicts require cost‐effective methods of monitoring wild animal behavior. Still and video camera surveillance can generate enormous quantities of data, which is laborious and expensive to screen for the species of interest. In the present study, we describe a state‐of‐the‐art, deep learning approach for automatically identifying and isolating species‐specific activity from still images and video data. We used a dataset consisting of 8,368 images of wild and domestic animals in farm buildings, and we developed an approach firstly to distinguish badgers from other species (binary classification) and secondly to distinguish each of six animal species (multiclassification). We focused on binary classification of badgers first because such a tool would be relevant to efforts to manage Mycobacterium bovis (the cause of bovine tuberculosis) transmission between badgers and cattle. We used two deep learning frameworks for automatic image recognition. They achieved high accuracies, in the order of 98.05% for binary classification and 90.32% for multiclassification. Based on the deep learning framework, a detection process was also developed for identifying animals of interest in video footage, which to our knowledge is the first application for this purpose. The algorithms developed here have wide applications in wildlife monitoring where large quantities of visual data require screening for certain species.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Deep Learning For Wildlife Species Recognitionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Wildlife surveillance using deep learning methods

Chen

Little

Mihaylova

et al. 2019

Ecology and Evolution

View full text Add to dashboard Cite

show abstract

“…A number of efforts are currently underway to test and refine automatic identification of species through state-ofthe-art machine learning algorithms, for example with deep convolutional neural networks (e.g. Nguyen et al 2017;Villa et al 2017;Norouzzadeh et al 2018;Tabak et al 2018). Fundamental to this will be an increased availability of open-access camera trap datasets (e.g.…”

Section: Number Of Responses Direction 54mentioning

confidence: 99%

Camera‐trapping version 3.0: current constraints and future priorities for development

Glover‐Kapfer

Soto-Navarro

Wearn

2019

Remote Sens Ecol Conserv

100

111

View full text Add to dashboard Cite

Camera traps are a widely used tool in wildlife research and conservation, but in situ factors such as theft, poor performance in extreme environments and damage by wildlife may be hindering the effectiveness of the technology. However, we still know little about how widespread these constraints are and which are the priorities to solve in the short‐ and mid‐term. We present results from a global survey of camera‐trappers working across a diversity of institutions and habitats, and using camera traps for a range of purposes. We show that: (1) the major current constraints on effective camera‐trapping are cost, theft and sensor performance (with 66%, 50% and 42% of respondents respectively classifying those as important or extremely important barriers for camera‐trapping); (2) the most‐needed technological developments are related to sensor performance (faster triggering responses and higher sensitivity), resistance to extreme environmental conditions (extreme temperatures and high humidity) and automated filtering of blank images; and (3) there is considerable variation among camera trap manufacturers in user‐rated performance, and none of the manufacturer ratings exhibited a trend over time, despite improvements in the technology. Our results serve as valuable market research for both open‐source and commercial camera trap developers. On the basis of our survey of camera‐trappers, we foresee a transition towards camera‐trapping 3.0 in the near‐future, consisting of both more effective camera trap units, but also greater use of wireless data transmission, sensor networks, automation of processes using algorithms and better, more collaborative tools for managing and analysing camera trap data. Ultimately, this will increase the capacity of researchers and conservationists to implement coordinated wildlife monitoring at unprecedented scales.

show abstract