Location Invariant Animal Recognition Using Mixed Source Datasets and Deep Learning

Abstract

1.A time-consuming challenge faced by camera trap practitioners all over the world is the extraction of meaningful data from images to inform ecological management. The primary methods of image processing used by practitioners includes manual analysis and citizen science. An increasingly popular alternative is automated image classification software. However, most automated solutions are not sufficiently robust to be deployed on a large scale. Key challenges include limited access to images for each species and lack of location invariance when transferring models between sites. This prevents optimal use of ecological data and results in significant expenditure of time and resources to annotate and retrain deep learning models.2. In this study, we aimed to (a) assess the value of publicly available non-iconic FlickR images in the training of deep learning models for camera trap object detection, (b) develop an out-ofthe-box location invariant automated camera trap image processing solution for ecologist using deep transfer learning and (c) explore the use of small subsets of camera trap images in optimisation of a FlickR trained deep learning model for high precision ecological object detection.3. We collected and annotated a dataset of images of "pigs" (Sus scrofa and Phacochoerus africanus) from the consumer image sharing website FlickR. These images were used to achieve transfer learning using a RetinaNet model in the task of object detection. We compared the performance of this model to the performance of models trained on combinations of camera trap images obtained from five different projects, each characterised by 5 different geographical regions. Furthermore, we explored optimisation of the FlickR model via infusion of small subsets of camera trap images to increase robustness in difficult images. 4. In most cases, the mean Average Precision (mAP) of the FlickR trained model when tested on out of sample camera trap sites (67.21-91.92%) was significantly higher than the mAP achieved by models trained on only one geographical location (4.42-90.8%) and rivalled the mAP of models trained on mixed camera trap datasets (68.96-92.75%). The infusion of camera trap images into the FlickR training further improved AP by 5.10-22.32% to 83.60-97.02%. Ecology researchers can use FlickR images in the training of automated deep learningsolutions for camera trap image processing to significantly reduce time and resource expenditure by allowing the development of location invariant, highly robust out-of-the-box solutions. This would allow AI technologies to be deployed on a large scale in ecological applications.

U-Infuse: Democratization of Customizable AI for Object Detection

“…If the results attained by the default models are not sufficiently accurate for the user's purposes, or the models are not sufficiently specific to the user's domain, they may elect to custom train their own model or models. They may choose to use publicly available images such as FlickR and iNaturalist (FiN) images, infused with camera trap images as proposed by (Shepley, Falzon et al 2020) or they may alternatively train using images from any source, including publicly available images, and/or their own trap images.…”

“…U-Infuse is an open source application which implements the location invariance methodology proposed by (Shepley, Falzon et al 2020). It democratises deep learning and AI technologies by making deep learning more accessible for ecological practitioners.…”

“…Furthermore, U-Infuse may be extended to support other frameworks including YOLOv3 (Redmon and Farhadi 2016), Faster RCNN (Ren, He et al 2015) and Single Shot Detectors (Zhang, Wen et al 2018). Incorporating FlickR API support to allow image downloading and sorting within U-Infuse, and integration of the SSIM image similarity tool and duplicate remover (Shepley, Falzon et al 2020) would also extend its usability in the development of high performance, domain specific deep learning object detectors. Copying, distribution and modification of U-Infuse source code is encouraged.…”

“…This need is addressed by U-Infuse, which provides a means by which ecological practitioners can easily train their own deep learning object detectors without requiring significant investment of time in the technical programming and specialised artificial intelligence domain knowledge required to develop custom models. U-Infuse is designed to enable ecologists to achieve location invariant object detection on camera trap images according to the methodology proposed by (Shepley, Falzon et al 2020). This would allow practitioners to process image data in-house, at their own pace according to their needs, removing the need to allocate significant time and resources to upload, save and share data with service providers which is a drawback of existing solutions.…”

Pest Animal's Detection and Habitat Identification in Low-resolution Airborne Thermal Imagery

“…However, manual detection of pest animals, habitat identification and estimation of pest population size is cumbersome as it requires frame by frame analysis of hours of video data. Some automated approaches are proposed in recent years [2,6,19,25,31]. However, they often lack usability due to low accuracy, ineffectiveness against occlusion, limitations of the visible spectrum and low detection speed.…”