Automatically detecting the wild giant panda using deep learning with context and species distribution model

The giant panda is a flagship species in ecological conservation. The infrared camera trap is an effective tool for monitoring the giant panda. Images captured by infrared camera traps must be accurately recognized before further statistical analyses can be implemented. Previous research has demonstrated that spatiotemporal and positional contextual information and the species distribution model (SDM) can improve image detection accuracy, especially for difficult-to-see images. Difficult-to-see images include those in which individual animals are only partially observed and it is challenging for the model to detect those individuals. By utilizing the attention mechanism, we developed a unique method based on deep learning that incorporates object detection, contextual information, and the SDM to achieve better detection performance in difficult-to-see images. We obtained 1169 images of the wild giant panda and divided them into a training set and a test set in a 4:1 ratio. Model assessment metrics showed that our proposed model achieved an overall performance of 98.1% in mAP_0.5 and 82.9% in recall on difficult-to-see images. Our research demonstrated that the fine-grained multimodal-fusing method applied to monitoring giant pandas in the wild can better detect the difficult-to-see panda images to enhance the wildlife monitoring system.     

基于双模型融合的大熊猫头部图像分割

在大熊猫(Ailuropoda melanoleuca)的迁地保护和种群饲养管理中,及时、快速地进行个体识别和行为监测,对其健康管理具有至关重要的作用。圈养大熊猫健康状况通常由专门的饲养人员肉眼观测,人力成本高、效率低并且缺乏时效性。基于图像的动物个体识别与行为分析技术效率高、时间成本低,已经成为新的监测发展趋势。已有研究提出,通过大熊猫面部图像的检测和分析,可实现个体识别和行为分类。但该方法依然存在检测精度不足导致识别准确率难以提升的问题。本文提出一种基于YOLOv3和Mask R-CNN的双模型融合方法,实现了大熊猫头部图像分割和精准检测。包含3个部分：YOLOv3完成头部检测,Mask R-CNN完成大熊猫轮廓分割,然后将两个模型的输出进行交并比融合。结果显示,头部检测准确率为82.6%,大熊猫轮廓分割准确率为95.2%,总体头部轮廓分割准确率为87.1%。该方法对大熊猫头部图像的识别率和分割准确率高,为大熊猫的个体识别、性别分类提供了帮助,为行为分析提供了技术参考。     

Black and white and read all over: the past,present and future of giant panda genetics

Few species attract much more attention from the public and scientists than the giant panda (Ailuropoda melanoleuca), a popular, enigmatic but highly endangered species. The application of molecular genetics to its biology and conservation has facilitated surprising insights into the biology of giant pandas as well as the effectiveness of conservation efforts during the past decades. Here, we review the history of genetic advances in this species, from phylogeny, demographical history, genetic variation, population structure, noninvasive population census and adaptive evolution to reveal to what extent the current status of the giant panda is a reflection of its evolutionary legacy, as opposed to the influence of anthropogenic factors that have negatively impacted this species. In addition, we summarize the conservation implications of these genetic findings applied for the management of this high‐profile species. Finally, on the basis of these advances and predictable future changes in genetic technology, we discuss future research directions that seem promising for giant panda biology and conservation.     

基于高分辨率网络的大熊猫姿态估计方法

对圈养大熊猫 (Ailuropoda melanoleuca) 开展长期行为监测能及时了解其所处生理周期和健康状况,有助于繁殖饲养机构迅速采取相应繁育保护措施提高饲养管理水平,但目前无法对大熊猫进行24 h监控并及时地获得相应的行为信息。准确的动物姿态估计是动物行为研究的关键,也是诸多下游应用的基础。了解大熊猫的姿态可以促进大熊猫行为研究并提升保护管理水平。为了提高复杂环境下大熊猫姿态估计的准确率,本文以高分辨率网络 (High resolution net, HRNet) 为基础网络架构提出了一种大熊猫姿态估计方法：针对大熊猫不同部位尺度差异较大的问题,在HRNet-32中引入了空洞空间金字塔池化 (Atrous spatial pyramid pooling, ASPP) 模块,在提升特征感受野的同时捕获多尺度信息;同时对大熊猫身体关键点进行分组,引入基于部位的多分支结构来学习特定于每个部位组的表征。多次对比实验结果表明本文所用模型具有较高的检测精度：在PCK@0.05中所用模型精度达到了81.51%。本文提出的方法可为大熊猫的行为分析和健康评估提供技术支撑。     

Gut microbiota in reintroduction of giant panda

Reintroduction is a key approach in the conservation of endangered species. In recent decades, many reintroduction projects have been conducted for conservation purposes, but the rate of success has been low. Given the important role of gut microbiota in health and diseases, we questioned whether gut microbiota would play a crucial role in giant panda's wild‐training process. The wild procedure is when captive‐born babies live with their mothers in a wilderness enclosure and learn wilderness survival skills from their mothers. During the wild‐training process, the baby pandas undergo wilderness survival tests and regular physical examinations. Based on their performance through these tests, the top subjects (age 2–3 years old) are released into the wild while the others are translocated to captivity. After release, we tracked one released panda (Zhangxiang) and collected its fecal samples for 5 months (January 16, 2013 to March 29 2014). Here, we analyzed the Illumina HiSeq sequencing data (V4 region of 16S rRNA gene) from captive pandas (n = 24), wild‐training baby pandas (n = 8) of which 6 were released and 2 were unreleased, wild‐training mother pandas (n = 8), one released panda (Zhangxiang), and wild giant pandas (n = 18). Our results showed that the gut microbiota of wild‐training pandas is significantly different from that of wild pandas but similar to that of captive ones. The gut microbiota of the released panda Zhangxiang gradually changed to become similar to those of wild pandas after release. In addition, we identified several bacteria that were enriched in the released baby pandas before release, compared with the unreleased baby pandas. These bacteria include several known gut‐health related beneficial taxa such as Roseburia, Coprococcus, Sutterella, Dorea, and Ruminococcus. Therefore, our results suggest that certain members of the gut microbiota may be important in panda reintroduction.     

Old-growth forest is what giant pandas really need

Giant pandas (Ailuropoda melanoleuca) are an iconic conservation species, but despite significant research effort, do we understand what they really need? Estimating and mapping suitable habitat play a critical role in conservation planning and policy. But if assumptions about ecological needs are wrong, maps with misidentified suitable habitat will misguide conservation action. Here, we use an information-theoretic approach to analyse the largest, landscape-level dataset on panda habitat use to date, and challenge the prevailing wisdom about panda habitat needs. We show that pandas are associated with old-growth forest more than with any ecological variable other than bamboo. Other factors traditionally used in panda habitat models, such as topographic slope, are less important. We suggest that our findings are disparate from previous research in part because our research was conducted over a larger ecological scale than previous research conducted over more circumscribed areas within individual reserves. Thus, extrapolating from habitat studies on small scales to conservation planning on large scales may entail some risk. As the Chinese government is considering the renewal of its logging ban, it should take heed of the panda''s dependency on old growth.     

野生大熊猫种群数量调查方法研究进展

大熊猫(Ailuropodamelanoleuca)是中国特有珍稀野生动物,被誉为野生动物保护领域的"旗舰物种"。大熊猫种群数量的变化可以实时反应大熊猫种群动态,提供大熊猫分布区域、栖息地质量等最直接的信息,是制定大熊猫保护方案的基础,也是有效实施大熊猫保护措施的前提。综述了几种传统野生大熊猫种群数量调查方法(包括直接计数法、数学模型法、距离-咬节分析法及分子生物学方法),以及近年来最新应用于野生动物种群数量调查的红外相机技术、足迹鉴定法,讨论了传统方法中可能存在的问题、分析新方法的应用前景,并针对今后的野生大熊猫种群数量调查提出了一些建议与展望。     

野生大熊猫现状、面临的挑战及展望

截至2003年底,我国野生大熊猫种群数量达1596只,分布在陕西、四川和甘肃3省的45个县境内,总栖息地面积达2304991hm^2。与第2次大熊猫调查相比,野生大熊猫生存状况已得到改善,分布范围扩大、栖息地面积增加、种群数量进一步增长。本文在第3次大熊猫调查的基础上,就野生大熊猫种群及栖息地现状进行了分析,指出未来保护大熊猫所面临的3个方面的挑战,即来自物种自身生物学特性的挑战、栖息地破碎化及隔离小种群未来命运的挑战以及大熊猫保护与社区经济发展需求相冲突的挑战。作者还就我国大熊猫保护前景进行了展望,即自然保护区数量将进一步增加,栖息地状况将进一步改善;种群数量在总体保持稳定的基础上将逐步增长,但局部小种群灭绝风险将加剧;圈养种群将形成能自我维持的种群,圈养个体通过培训将逐步放归到隔离野生小种群中以改变其命运。     

汶川地震对大熊猫栖息地的影响与恢复对策

 大熊猫是生物多样性保护的旗舰种, 保护大熊猫及其栖息地是保护生物多样性和生态系统功能完整性与稳定性的重要保障体现。汶川地震灾区位于大熊猫重点分布区岷山-邛崃山, 地震及其次生灾害导致该区27个大熊猫自然保护区不同程度受损, 8.3%的大熊猫栖息地因地震而被破坏。地震及其次生灾害对大熊猫栖息地的影响主要表现在: 1) 地震埋没和砸毁大熊猫赖以生存的主食竹, 地震可能诱发主食竹开花, 威胁到大熊猫的健康和食物安全; 2) 地震及其诱发的土壤和山石运动显著影响森林的动态特征, 森林大面积丧失或质量下降; 3) 地震改变大熊猫栖息地生境特征, 大熊猫个体交流的廊道阻断, 形成“生殖孤岛”, 遗传多样性降低, 栖息地破碎化进程加快。应对震后大熊猫栖息地恢复的主要对策有: 1) 重新评估震后大熊猫栖息地质量, 并重新规划现有大熊猫保护区群的布局; 2) 应用地理信息系统、遥感及数学模型等手段与野外实地实证研究相结合的方法, 全面查清震后大熊猫栖息地主食竹资源状况及分布规律并及时监测其动态, 复壮更新大熊猫主食竹; 3) 利用天然植被自然恢复和人工重建等措施恢复因地震而退化或丧失的大熊猫栖息地。     

Impacts of temperature on giant panda habitat in the north Minshan Mountains

Understanding the impacts of meteorological factors on giant pandas is necessary for future conservation measures in response to global climate change. We integrated temperature data with three main habitat parameters (elevation, vegetation type, and bamboo species) to evaluate the influence of climate change on giant panda habitat in the northern Minshan Mountains using a habitat assessment model. Our study shows that temperature (relative importance = 25.1%) was the second most important variable influencing giant panda habitat excepting the elevation. There was a significant negative correlation between temperature and panda presence (ρ = −0.133, P < 0.05), and the temperature range preferred by giant pandas within the study area was 18–21°C, followed by 15–17°C and 22–24°C. The overall suitability of giant panda habitats will increase by 2.7%, however, it showed a opposite variation patterns between the eastern and northwestern region of the study area. Suitable and subsuitable habitats in the northwestern region of the study area, which is characterized by higher elevation and latitude, will increase by 18007.8 hm² (9.8% habitat suitability), while the eastern region will suffer a decrease of 9543.5 hm² (7.1% habitat suitability). Our results suggest that increasing areas of suitable giant panda habitat will support future giant panda expansion, and food shortage and insufficient living space will not arise as problems in the northwest Minshan Mountains, which means that giant pandas can adapt to climate change, and therefore may be resilient to climate change. Thus, for the safety and survival of giant pandas in the Baishuijiang Reserve, we propose strengthening the giant panda monitoring program in the west and improving the integrity of habitats to promote population dispersal with adjacent populations in the east.     

Application of ecological-niche factor analysis in habitat assessment of giant pandas

Ecological-niche factor analysis (ENFA) is a multivariate approach to study geographic distribution of species on a large scale with only “presence” data. It has been widely applied in many fields including wildlife management, habitat assessment and habitat prediction. In this paper, this approach was applied in habitat suitability assessment for giant pandas in Pingwu County, Sichuan Province, China. With “presence” data of giant pandas and remote sensing data, habitat suitability of pandas in this county was evaluated based on ENFA model, and spatial distribution pattern of nature reserves and conservation gaps were then evaluated. The results show that giant pandas in this county prefer high-elevation zones (> 2128 m) dominated by coniferous forest, and mixed coniferous and deciduous broadleaf forest, and avoid deciduous broadleaf forest and shrubs. Pandas avoid staying at habitats with human disturbances. Farmland is a major factor threatening panda habitat. Panda habitat is mainly distributed in north and west of Pingwu with a total area of 234033 hm², 106345hm² for suitable habitat and 127688 hm² for marginally suitable habitat). 3 nature reserves were located in Pingwu, covering over 49.2% of total suitable habitat and 45.6% of total marginally suitable habitat. Although 47.2% of panda habitat was in reserves under protection, connectivity between reserves was weak and a conservation gap existed in the north part of Pingwu. Thus, a new nature reserve in Baima and Mupi should be established to link the isolated habitats.     

四川省大熊猫保护地生态安全及其时空演变

大熊猫是生态保护的旗舰物种,保护大熊猫及其栖息地具有重要的示范作用和科学价值。中国致力于建设以国家公园为主体的自然保护地体系,分析大熊猫保护地生态安全,为建立大熊猫国家公园提供理论依据,具有重要意义。以四川省为例,选取被划入或将来可能被划入大熊猫保护地的典型县域,构建大熊猫保护地生态安全评价指标体系,采用熵权法、层次分析法进行综合评价,并结合空间相关性分析,从时间和空间两个维度对大熊猫保护地生态安全状况进行时空动态分析。结果表明：（1）2003-2017年四川省大熊猫保护地生态安全综合指数呈波动性上升趋势,生态环境改善显著,但仍有部分县域出现不同程度的环境恶化现象。（2）各县域生态安全状况表现出较强的差异性,并且相邻县域生态安全指数存在显著的空间正相关性和空间集聚性。（3）大熊猫保护地生态安全与周边社会经济发展之间矛盾依然尖锐,相邻县域之间的经济发展和生态保护状况相互影响的作用明显,需联合治理,共同促进生态平衡,提高大熊猫及其栖息地保护效果。     

Preliminary investigation on iodine nutrition in captive giant pandas

Background/ObjectiveThe giant panda belongs to the family Ursidae and, as a species of bear, still retains the simple digestive system of a Carnivoran. However, under the pressure of a specific habitat they had to adapt to a plant mono-diet consisting of bamboo with different species and growth stages around the year. A plant-based diet has relatively low iodine content with risk of iodine deficiency. Furthermore, bamboo contains cyanogenic glycosides releasing cyanide whose detoxification metabolite the thiocyanate acts as antagonist against iodine uptake and storage in the thyroid. To date very little is known about the iodine nutritional status of the giant panda, thus this study was conducted to receive the first information about the iodine nutrition of captive giant panda.Subjects/MethodsHere we investigated the iodine content of bamboo with different plant parts/vegetation stage and species and further compounds of the captive giant panda diet. Next, the urinary iodine (UI) and urinary thiocyanate (UT) levels of infant, sub-adult, adult and geriatric captive giant pandas was measured during the periods when the pandas consume both bamboo leaves- and culm (bamboo leaf-culm stage). Afterwards, the UI of 19 adult giant pandas was measured again for the different iodine intake during bamboo shoot stage. Finally, in this study part also the fecal iodine concentration was analyzed for calculation of total iodine excretion in relation to the iodine intake.ResultsBamboo leaves had the highest iodine content (453 μg/kg dry matter (DM)), followed by the shoots (84 μg/kg DM, p < 0.05), while bamboo culm had the lowest value (12 μg/kg DM, p < 0.05). During bamboo leaf-culm stage, giant pandas of different age groups had different UI and UT levels (p < 0.05). Furthermore, UI and UT were positively correlated among sub-adult, adult and geriatric giant pandas (p < 0.05). In adult giant pandas during bamboo shoot stage, the iodine excretion in feces was not different from that in urine while their total iodine excretion was less than their iodine intake (p < 0.05). Moreover, during bamboo shoot stage, the UI level of adult giant pandas was much lower than noted during bamboo leaf-culm stage (p < 0.05).ConclusionsOur results indicate that UI of captive giant pandas was related to their age as well as to the vegetation stage/part of bamboo they consumed reflecting a different periodic iodine supply. Thiocyanate and fecal excretion should be emphasized when considering the iodine nutrition of giant pandas.     

Giant pandas are losing their edge: Population trend and distribution dynamic drivers of the giant panda

Comprehending the population trend and understanding the distribution range dynamics of species are necessary for global species protection. Recognizing what causes dynamic distribution change is crucial for identifying species' environmental preferences and formulating protection policies. Here, we studied the rear-edge population of the flagship species, giant pandas (Ailuropoda melanoleuca), to (1) assess their population trend using their distribution patterns, (2) evaluate their distribution dynamics change from the second (1988) to the third (2001) survey (2–3 Interval) and third to the fourth (2013) survey (3–4 Interval) using a machine learning algorithm (eXtremely Gradient Boosting), and (3) decode model results to identify driver factors in the first known use of SHapley Additive exPlanations. Our results showed that the population trends in Liangshan Mountains were worst in the second survey (k = 1.050), improved by the third survey (k = 0.97), but deteriorated by the fourth survey (k = 0.996), which indicates a worrying population future. We found that precipitation had the most significant influence on distribution dynamics among several potential environmental factors, showing a negative correlation between precipitation and giant panda expansion. We recommend that further research is needed to understand the microenvironment and animal distribution dynamics. We provide a fresh perspective on the dynamics of giant panda distribution, highlighting novel focal points for ecological research on this species. Our study offers theoretical underpinnings that could inform the formulation of more effective conservation policies. Also, we emphasize the uniqueness and importance of the Liangshan Mountains giant pandas as the rear-edge population, which is at a high risk of population extinction.     

大熊猫与同域动物在海拔分布上的生态位分化

大熊猫（Ailuropoda melanoleuca）是中国的特有物种,其成功放归可能受放归地气候、栖息地、同域动物等多种因素的影响。本研究在野外调查的基础上,基于生态位理论等方法,探讨了栗子坪国家级自然保护区大熊猫与其大中型同域动物的空间关系。调查结果表明,该保护区内大熊猫同域动物共20种,分别属于2纲5目,其中东洋型分布物种占优势（30%）。大熊猫同域动物分布海拔显著低于大熊猫的分布海拔（P＜0.05）,大熊猫分别与灵长目等5个目的动物在海拔分布上均存在显著的生态位分化（P＜0.001）。灵长目动物在空间分布上与大熊猫分布相异,而其他目的动物与大熊猫空间分布类似。偶蹄目动物痕迹数量占比最高（54.14 %）,而食肉目的小熊猫与大熊猫样线共同遇见率达到了43.75%。本研究结果表明,该地大熊猫同域动物较为丰富,在海拔分布上与大熊猫存在生态位分化。该研究可为圈养大熊猫的放归提供参考依据。     

气候变化对邛崃山系大熊猫主食竹和栖息地分布的影响

气候变化对生物多样性的影响,特别是珍稀濒危物种的影响是当前的研究热点。全球气候变化对大熊猫的影响一直受到广泛关注。根据野外调查的大熊猫活动痕迹点、竹类分布点和主食竹扩散距离数据,采用Maxent模型,利用植被、地形、气候等因素,在RCP8.5下分析了2050年和2070年邛崃山系大熊猫主食竹分布及栖息地变化趋势。结果显示:(1)未来大熊猫适宜生境及主食竹气候适宜区面积均有所减少,到2070年分别减少37.2%和4.7%;(2)未来主食竹分布范围总体向高海拔扩展,但面积持续减少,到2070年分布面积比当前减少8.3%;(3)大熊猫栖息地未来有向高海拔扩张的趋势,在低海拔地区退缩明显,到2070年较当前减少27.2%;但到2070年大熊猫栖息地面积加上非栖息地有主食竹分布的面积,较现有大熊猫栖息地面积大1.5%;(4)受气候变化影响较严重的区域是邛崃山系南部以及低海拔地区,其余区域所受影响相对较小;(5)未来需要加强对受气候变化影响严重区域的监测与保护,特别是邛崃山系中部的大熊猫集中分布区。     

Habitat Assessment for Giant Pandas in the Qinling Mountain Region of China

ABSTRACT Because habitat loss and fragmentation threaten giant pandas (Ailuropoda melanoleuca), habitat protection and restoration are important conservation measures for this endangered species. However, distribution and value of potential habitat to giant pandas on a regional scale are not fully known. Therefore, we identified and ranked giant panda habitat in Foping Nature Reserve, Guanyinshan Nature Reserve, and adjacent areas in the Qinling Mountains of China. We used Mahalanobis distance and 11 digital habitat layers to develop a multivariate habitat signature associated with 247 surveyed giant panda locations, which we then applied to the study region. We identified approximately 128 km²of giant panda habitat in Foping Nature Reserve (43.6% of the reserve) and 49 km²in Guanyinshan Nature Reserve (33.6% of the reserve). We defined core habitat areas by incorporating a minimum patch-size criterion (5.5 km²) based on home-range size. Percentage of core habitat area was higher in Foping Nature Reserve (41.8% of the reserve) than Guanyinshan Nature Reserve (26.3% of the reserve). Within the larger analysis region, Foping Nature Reserve contained 32.7% of all core habitat areas we identified, indicating regional importance of the reserve. We observed a negative relationship between distribution of core areas and presence of roads and small villages. Protection of giant panda habitat at lower elevations and improvement of habitat linkages among core habitat areas are important in a regional approach to giant panda conservation.     

Wildlife surveillance using deep learning methods

相似文献   

基于发育网络识别模型的大熊猫面部识别

个体识别是动物行为学与生态学研究工作的基础,也是制定珍稀野生动物保护政策的重要依据。为了丰富大熊猫个体识别和种群数量调查的方法,我们于2017年7月分别在四川省雅安市碧峰峡大熊猫基地和四川省汶川县耿达镇的中华大熊猫苑共计拍摄18只大熊猫个体,每只大熊猫拍摄6~13张高质量面部照片（共计131张）,利用发育网络（Developmental Network）建立大熊猫面部识别模型。利用此模型对存在部分背景的大熊猫面部照片进行识别检测,得到的个体识别率为79.41%,对完全去除背景的大熊猫面部照片进行识别检测,得到的个体识别率为58.82%。研究表明,发育网络具有足够的大熊猫个体识别能力,不同背景比例的照片对大熊猫个体识别的实际结果具有较大的影响。随着发育网络识别模型的发展,我们建议更多的野生动物保护研究者结合这一技术深入地开展珍稀野生动物（如大熊猫）个体识别研究,逐步提高识别准确度,并应用到关键区域大规模的动物调查中。     

Parasite Threat to Panda Conservation

The giant panda is a global symbol of wildlife conservation that is threatened by historic and current habitat loss. Despite a great deal of research on the physiology, reproductive biology, and diet of pandas in the wild and in captivity, there is little information on wild panda mortality. Here we integrate previously unavailable data on the mortality of wild pandas. We report on three recent phases of panda mortality: deaths due to bamboo flowering in the 1970s and 1980s, surprisingly extensive poaching in the 1980s and 1990s, and a parasitic infection over the past few years. Our analyses suggest that the current most significant threat to wild panda survival is disease due to extraintestinal migration (visceral larval migrans) by an ascarid nematode. We demonstrate that the probability of death of wild pandas being caused by this disease increased significantly between 1971 and 2005 and discuss the possible factors leading to the emergence of this disease. Electronic Supplementary Material The online version of this article (doi:) contains supplementary material, which is available to authorized users.