Automated face detection for occurrence and occupancy estimation in chimpanzees

Pre-trained models are not enough: active and lifelong learning is important for long-term visual monitoring of mammals in biodiversity research—Individual identification and attribute prediction with image features from deep neural networks and decoupled decision models applied to elephants and great apes

Animal re-identification based on image data, either recorded manually by photographers or automatically with camera traps, is an important task for ecological studies about biodiversity and conservation that can be highly automatized with algorithms from computer vision and machine learning. However, fixed identification models only trained with standard datasets before their application will quickly reach their limits, especially for long-term monitoring with changing environmental conditions, varying visual appearances of individuals over time that differ a lot from those in the training data, and new occurring individuals that have not been observed before. Hence, we believe that active learning with human-in-the-loop and continuous lifelong learning is important to tackle these challenges and to obtain high-performance recognition systems when dealing with huge amounts of additional data that become available during the application. Our general approach with image features from deep neural networks and decoupled decision models can be applied to many different mammalian species and is perfectly suited for continuous improvements of the recognition systems via lifelong learning. In our identification experiments, we consider four different taxa, namely two elephant species: African forest elephants and Asian elephants, as well as two species of great apes: gorillas and chimpanzees. Going beyond classical re-identification, our decoupled approach can also be used for predicting attributes of individuals such as gender or age using classification or regression methods. Although applicable for small datasets of individuals as well, we argue that even better recognition performance will be achieved by improving decision models gradually via lifelong learning to exploit huge datasets and continuous recordings from long-term applications. We highlight that algorithms for deploying lifelong learning in real observational studies exist and are ready for use. Hence, lifelong learning might become a valuable concept that supports practitioners when analyzing large-scale image data during long-term monitoring of mammals.

Research on Face Recognition Algorithm Based on Image Processing

While network technology is convenient for our daily life, the problems that are exposed are also endless. The most important thing for everyone is information security. In order to improve the security level of network information and identify and detect faces, the method used in this paper has improved compared with the traditional AdaBoost method and skin color method. AdaBoost detection is performed on the image, which reduces the probability of false detection. The experiment compares the experimental results of the AdaBoost method, the skin color method and the skin color + AdaBoost method. All operations in the KPCA and KFDA algorithms are performed by the inner product kernel function defined in the original space, and no specific non-linear mapping function is involved.The full name of KPCA is kernel principal component analysis. The full name of KFDA is kernel Fisher discriminant analysis. Combining the zero-space method kernel discriminant analysis method improves the ability of discriminant analysis to extract non-linear features. Through the secondary extraction of PCA features, a better recognition result than the PCA method is obtained. This paper also proposes a zero-space based Fisher discriminant analysis method. Experiments show that the zero-space-based method makes full use of the useful discriminant information in the zero space of the intraclass dispersion matrix, which improves the accuracy of face recognition to some extent.If you choose the polynomial kernel function, when d = 0.8, KPCA has a higher recognition ability. When d = 2, the recognition rate of KFDA and zero space-based KFDA is the largest. For polynomial functions, in general, d = 2.

Noninvasive Technologies for Primate Conservation in the 21st Century

Observing and quantifying primate behavior in the wild is challenging. Human presence affects primate behavior and habituation of new, especially terrestrial, individuals is a time-intensive process that carries with it ethical and health concerns, especially during the recent pandemic when primates are at even greater risk than usual. As a result, wildlife researchers, including primatologists, have increasingly turned to new technologies to answer questions and provide important data related to primate conservation. Tools and methods should be chosen carefully to maximize and improve the data that will be used to answer the research questions. We review here the role of four indirect methods—camera traps, acoustic monitoring, drones, and portable field labs—and improvements in machine learning that offer rapid, reliable means of combing through large datasets that these methods generate. We describe key applications and limitations of each tool in primate conservation, and where we anticipate primate conservation technology moving forward in the coming years.

Using citizen science in the photo-identification of adult individuals of an amphibian based on two facial skin features

Among amphibians, adults have traditionally been identified in capture-mark-recapture studies using invasive marking techniques with associated ethical, cost and logistical considerations. However, species in this group may be strong candidates for photo-identification based on natural skin features that removes many of these concerns, with this technique opening up opportunities for citizen scientists to be involved in animal monitoring programs. We investigated the feasibility of using citizen science to distinguish between individuals of an Australian anuran (the sandpaper frog, Lechriodus fletcheri) based on a visual analysis of their natural skin features. We collected photographs of marked individuals in the field over three breeding seasons using a smartphone device. This photo-database was used to create an online survey to determine how easily members of the general public could photo-match individuals by a comparison of two facial skin features; black banding that runs horizontally above the tympanum and a background array of tubercles present in this region. Survey participants were provided with 30 closed, multiple choice questions in which they were asked to match separate images of a query frog from small image pools of potential candidate matches. Participants were consistently able to match individuals with a low matching error rate (mean ± SD of 26 ± 5) despite the relatively low quality of photographs taken from a smartphone device in the field, with most query frogs being matched by a majority of participants (mean ± SD of 86.02 ± 9.52%). These features were found to be unique and stable among adult males and females. Thus, photo-identification is likely to be a valid, non-invasive method for capture-mark-recapture for L. fletcheri, and likely many anurans that display similar facial skin features. This may become an important alternative to artificial marking techniques, with the challenges of manual photo-matching reduced by spreading workloads among members of the public that can be recruited online.

Optimising deployment time of remote cameras to estimate abundance of female bighorn sheep

Abstract ContextWildlife biologists accumulate large quantities of images from remote cameras, which can be time- and cost-prohibitive to archive and analyse. Remote-camera projects would benefit from not setting cameras longer than needed and not analysing more images than needed; however, there is a lack of information about optimal deployment time required for remote-camera surveys to estimate ungulate abundance. AimsThe objective was to estimate abundance of adult females in a population of Rocky Mountain bighorn sheep (Ovis canadensis canadensis) in Utah, USA, from 2012 to 2014, and determine whether this type of study can be conducted more efficiently. Because females are the most important cohort for population growth, remote cameras were set at three water sources and mark–resight models in Program MARK were used. MethodsWe compared estimated abundance of collared and uncollared females by number of days cameras were set using 31 replicated abundance estimates from each year starting 1 July. Each replicated estimate used a different number of days and photographs from a 62-day sampling period (1 July to 31 August). Key resultsAbundance estimates ranged from 44 to 98 animals. Precise estimates of abundance, however, were obtained with only 12 days of sampling in each year. By analysing only 12 days of images rather than 62 days in all years, the estimated mean of 58 adult females would have changed by only 7 individuals (±4 individuals, range=3–10 animals), the s.e. would have increased by a mean of only 4 individuals (±1.6, range=2.0–5.2 individuals) and a mean of only 18% (±10.5%, range=8–29%) of images would have been analysed. Across the study, analysis of >23000 (>80%) images could have been avoided, saving time and money. ConclusionsThe results indicate that an asymptotic relationship exists between estimated abundance of female bighorn sheep and remote-camera deployment time. ImplicationsThe mark–resight methods used in the present study would work for other ungulates in which individuals are radio collared or marked using remote cameras set at water sources, trail crossings or mineral licks. These findings can help researchers reduce cost of setting, servicing, archiving and analysing photographs from remote cameras for ungulate population monitoring.

Comparing Methods for Assessing Chimpanzee (Pan troglodytes schweinfurthii) Party Size: Observations, Camera Traps, and Bed Counts from a Savanna–Woodland Mosaic in the Issa Valley, Tanzania

Studying animal grouping behavior is important for understanding the causes and consequences of sociality and has implications for conservation. Chimpanzee (Pan troglodytes) party size is often assessed by counting individuals or extracted indirectly from camera trap footage or the number of nests. Little is known, however, about consistency across methods for estimating party size. We collected party size data for wild chimpanzees in the Issa valley, western Tanzania, using direct observations, camera traps, and nest counts over six years (2012–2018). We compared mean monthly party size estimates calculated using each method and found that estimates derived from direct observations were weakly positively correlated with those derived from camera traps. Estimates from nest counts were not significantly correlated with either direct observations or camera traps. Overall observed party size was significantly larger than that estimated from both camera traps and nest counts. In both the dry and wet seasons, observed party size was significantly larger than camera trap party size, but not significantly larger than nest party size. Finally, overall party size and wet season party size estimated from camera traps were significantly smaller than nest party size, but this was not the case in the dry season. Our results reveal how data collection methods influence party size estimates in unhabituated chimpanzees and have implications for comparative analysis within and across primate communities. Specifically, future work must consider how estimates were calculated before we can reliably investigate environmental influences on primate behavior.

Validating camera trap distance sampling for chimpanzees

The extension of distance sampling methods to accommodate observations from camera traps has recently enhanced the potential to remotely monitor multiple species without the need of additional data collection (sign production and decay rates) or individual identification. However, the method requires that the proportion of time is quantifiable when animals can be detected by the cameras. This can be problematic, for instance, when animals spend time above the ground, which is the case for most primates. In this study, we aimed to validate camera trap distance sampling (CTDS) for the semiarboreal western chimpanzee (Pan troglodytes verus) in Taï National Park, Côte d'Ivoire by estimating abundance of a population of known size and comparing estimates to those from other commonly applied methods. We estimated chimpanzee abundance using CTDS and accounted for limited availability for detection (semiarboreal). We evaluated bias and precision of estimates, as well as costs and efforts required to obtain them, and compared them to those from spatially explicit capture-recapture (SECR) and line transect nest surveys. Abundance estimates obtained by CTDS and SECR produced a similar negligible bias, but CTDS yielded a larger coefficient of variation (CV = 39.70% for CTDS vs. 1%/19% for SECR). Line transects generated the most biased abundance estimates but yielded a better coefficient of variation (27.40-27.85%) than CTDS. Camera trap surveys were twice more costly than line transects because of the initial cost of cameras, while line transects surveys required more than twice as much time in the field. This study demonstrates the potential to obtain unbiased estimates of the abundance of semiarboreal species like chimpanzees by CTDS. HIGHLIGHTS: Camera trap distance sampling produced accurate density estimates for semiarboreal chimpanzees. Availability for detection must be accounted for and can be derived from the activity pattern.