Fine-grained Species Research Articles

Fine-grained Species Recognition with Privileged Pooling: Better Sample Efficiency Through Supervised Attention.

We propose a scheme for supervised image classification that uses privileged information, in the form of keypoint annotations for the training data, to learn strong models from small and/or biased training sets. Our main motivation is the recognition of animal species for ecological applications such as biodiversity modelling, which is challenging because of long-tailed species distributions due to rare species, and strong dataset biases such as repetitive scene background in camera traps. To counteract these challenges, we propose a visual attention mechanism that is supervised via keypoint annotations that highlight important object parts. This privileged information, implemented as a novel privileged pooling operation, is only required during training and helps the model to focus on regions that are discriminative. In experiments with three different animal species datasets, we show that deep networks with privileged pooling can use small training sets more efficiently and generalize better.

Sphere2Vec: A general-purpose location representation learning over a spherical surface for large-scale geospatial predictions

Generating learning-friendly representations for points in space is a fundamental and long-standing problem in machine learning. Recently, multi-scale encoding schemes (such as Space2Vec and NeRF) were proposed to directly encode any point in 2D or 3D Euclidean space as a high-dimensional vector, and has been successfully applied to various (geo)spatial prediction and generative tasks. However, all current 2D and 3D location encoders are designed to model point distances in Euclidean space. So when applied to large-scale real-world GPS coordinate datasets (e.g., species or satellite images taken all over the world), which require distance metric learning on the spherical surface, both types of models can fail due to the map projection distortion problem (2D) and the spherical-to-Euclidean distance approximation error (3D). To solve these problems, we propose a multi-scale location encoder called Sphere2Vec which can preserve spherical distances when encoding point coordinates on a spherical surface. We developed a unified view of distance-reserving encoding on spheres based on the Double Fourier Sphere (DFS). We also provide theoretical proof that the Sphere2Vec encoding preserves the spherical surface distance between any two points, while existing encoding schemes such as Space2Vec and NeRF do not. Experiments on 20 synthetic datasets show that Sphere2Vec can outperform all baseline models including the state-of-the-art (SOTA) 2D location encoder (i.e., Space2Vec) and 3D encoder NeRF on all these datasets with up to 30.8% error rate reduction. We then apply Sphere2Vec to three geo-aware image classification tasks - fine-grained species recognition, Flickr image recognition, and remote sensing image classification. Results on 7 real-world datasets show the superiority of Sphere2Vec over multiple 2D and 3D location encoders on all three tasks. Further analysis shows that Sphere2Vec outperforms other location encoder models, especially in the polar regions and data-sparse areas because of its nature for spherical surface distance preservation. Code and data of this work are available at https://gengchenmai.github.io/sphere2vec-website/.

Fine-Grained Butterfly Recognition via Peer Learning Network with Distribution-Aware Penalty Mechanism.

Simple SummaryAutomatic species recognition, such as butterflies or other insects, plays a crucial role in intelligent agricultural production management and the study of species diver-sity. However, the quite diverse and subtle interspecific differences and the long-tailed distribution of sample data in fine-grained species recognition are insufficient to learn robust feature representation and alleviate the bias and variance problems of the long-tailed classifier on insect recognition. The objective of this study is to develop a peer learning network with a distribution-aware penalty mechanism proposed to learn discriminative feature representation and mitigate the bias and variance problems in the long-tailed distribution. The results of various contrast experiments on collecting the butterfly-914 dataset show that the proposed PLN-DPM has a higher Rank-1 ac-curacy rate (86.2% on the butterfly dataset and 73.51% on the IP102 dataset). Addi-tionally, we deployed the PLN-DPM model on the smartphone app for butterfly recognition in a real-life environment.Automatic species recognition plays a key role in intelligent agricultural production management and the study of species diversity. However, fine-grained species recognition is a challenging task due to the quite diverse and subtle interclass differences among species and the long-tailed distribution of sample data. In this work, a peer learning network with a distribution-aware penalty mechanism is proposed to address these challenges. Specifically, the proposed method employs the two-stream ResNeSt-50 as the backbone to obtain the initial predicted results. Then, the samples, which are selected from the instances with the same predicted labels by knowledge exchange strategy, are utilized to update the model parameters via the distribution-aware penalty mechanism to mitigate the bias and variance problems in the long-tailed distribution. By performing such adaptive interactive learning, the proposed method can effectively achieve improved recognition accuracy for head classes in long-tailed data and alleviate the adverse effect of many head samples relative to a few samples of the tail classes. To evaluate the proposed method, we construct a large-scale butterfly dataset (named Butterfly-914) that contains approximately 72,152 images belonging to 914 species and at least 20 images for each category. Exhaustive experiments are conducted to validate the efficiency of the proposed method from several perspectives. Moreover, the superior Top-1 accuracy rate (86.2%) achieved on the butterfly dataset demonstrates that the proposed method can be widely used for agricultural species identification and insect monitoring.

Learning representation for multiple biological networks via a robust graph regularized integration approach.

Learning node representation is a fundamental problem in biological network analysis, as compact representation features reveal complicated network structures and carry useful information for downstream tasks such as link prediction and node classification. Recently, multiple networks that profile objects from different aspects are increasingly accumulated, providing the opportunity to learn objects from multiple perspectives. However, the complex common and specific information across different networks pose challenges to node representation methods. Moreover, ubiquitous noise in networks calls for more robust representation. To deal with these problems, we present a representation learning method for multiple biological networks. First, we accommodate the noise and spurious edges in networks using denoised diffusion, providing robust connectivity structures for the subsequent representation learning. Then, we introduce a graph regularized integration model to combine refined networks and compute common representation features. By using the regularized decomposition technique, the proposed model can effectively preserve the common structural property of different networks and simultaneously accommodate their specific information, leading to a consistent representation. A simulation study shows the superiority of the proposed method on different levels of noisy networks. Three network-based inference tasks, including drug-target interaction prediction, gene function identification and fine-grained species categorization, are conducted using representation features learned from our method. Biological networks at different scales and levels of sparsity are involved. Experimental results on real-world data show that the proposed method has robust performance compared with alternatives. Overall, by eliminating noise and integrating effectively, the proposed method is able to learn useful representations from multiple biological networks.

Efficient Deep Learning Models for Categorizing Chenopodiaceae in the Wild

The Chenopodiaceae species are ecologically and financially important, and play a significant role in biodiversity around the world. Biodiversity protection is critical for the survival and sustainability of each ecosystem and since plant species recognition in their natural habitats is the first process in plant diversity protection, an automatic species classification in the wild would greatly help the species analysis and consequently biodiversity protection on earth. Computer vision approaches can be used for automatic species analysis. Modern computer vision approaches are based on deep learning techniques. A standard dataset is essential in order to perform a deep learning algorithm. Hence, the main goal of this research is to provide a standard dataset of Chenopodiaceae images. This dataset is called ACHENY and contains 27030 images of 30 Chenopodiaceae species in their natural habitats. The other goal of this study is to investigate the applicability of ACHENY dataset by using deep learning models. Therefore, two novel deep learning models based on ACHENY dataset are introduced: First, a lightweight deep model which is trained from scratch and is designed innovatively to be agile and fast. Second, a model based on the EfficientNet-B1 architecture, which is pre-trained on ImageNet and is fine-tuned on ACHENY. The experimental results show that the two proposed models can do Chenopodiaceae fine-grained species recognition with promising accuracy. To evaluate our models, their performance was compared with the well-known VGG-16 model after fine-tuning it on ACHENY. Both VGG-16 and our first model achieved about 80% accuracy while the size of VGG-16 is about 16[Formula: see text] larger than the first model. Our second model has an accuracy of about 90% and outperforms the other models where its number of parameters is 5[Formula: see text] than the first model but it is still about one-third of the VGG-16 parameters.

Robust Species Distribution Mapping of Crop Mixtures Using Color Images and Convolutional Neural Networks.

Crop mixtures are often beneficial in crop rotations to enhance resource utilization and yield stability. While targeted management, dependent on the local species composition, has the potential to increase the crop value, it comes at a higher expense in terms of field surveys. As fine-grained species distribution mapping of within-field variation is typically unfeasible, the potential of targeted management remains an open research area. In this work, we propose a new method for determining the biomass species composition from high resolution color images using a DeepLabv3+ based convolutional neural network. Data collection has been performed at four separate experimental plot trial sites over three growing seasons. The method is thoroughly evaluated by predicting the biomass composition of different grass clover mixtures using only an image of the canopy. With a relative biomass clover content prediction of R2 = 0.91, we present new state-of-the-art results across the largely varying sites. Combining the algorithm with an all terrain vehicle (ATV)-mounted image acquisition system, we demonstrate a feasible method for robust coverage and species distribution mapping of 225 ha of mixed crops at a median capacity of 17 ha per hour at 173 images per hectare.

Bilinear pyramid network for flower species categorization

It is a challenging task to distinguish between numerous species of flowers due to their visually similarities and variations of the pose and structure. Thanks to properly modeling of the local feature interactions, bilinear CNN has succeeded in classifying of many non-rigid fine-grained species including flowers. However, bilinear CNN only computes the feature in a straightforward way without exploring the interactions between features from multiple layers in the network. In this paper, we present a novel Bilinear Pyramid Network (BPN) for flower categorization. Instead of passing through the network and directly feeding the final classifier, features from a convolutional layer are resized and multiplied with that from the former layer, which alternates multiple times to generates prediction vectors using the features from distinct layers. These features encoded from the feature pyramid spontaneously carry multi-level semantic cues, which yields stronger discriminative powers than single-layer features. Experiments show that the proposed network obtains superior classification results on the challenging dataset of flowers.

From Google Maps to a fine-grained catalog of street trees

Up-to-date catalogs of the urban tree population are of importance for municipalities to monitor and improve quality of life in cities. Despite much research on automation of tree mapping, mainly relying on dedicated airborne LiDAR or hyperspectral campaigns, tree detection and species recognition is still mostly done manually in practice. We present a fully automated tree detection and species recognition pipeline that can process thousands of trees within a few hours using publicly available aerial and street view images of Google MapsTM. These data provide rich information from different viewpoints and at different scales from global tree shapes to bark textures. Our work-flow is built around a supervised classification that automatically learns the most discriminative features from thousands of trees and corresponding, publicly available tree inventory data. In addition, we introduce a change tracker that recognizes changes of individual trees at city-scale, which is essential to keep an urban tree inventory up-to-date. The system takes street-level images of the same tree location at two different times and classifies the type of change (e.g., tree has been removed). Drawing on recent advances in computer vision and machine learning, we apply convolutional neural networks (CNN) for all classification tasks. We propose the following pipeline: download all available panoramas and overhead images of an area of interest, detect trees per image and combine multi-view detections in a probabilistic framework, adding prior knowledge; recognize fine-grained species of detected trees. In a later, separate module, track trees over time, detect significant changes and classify the type of change. We believe this is the first work to exploit publicly available image data for city-scale street tree detection, species recognition and change tracking, exhaustively over several square kilometers, respectively many thousands of trees. Experiments in the city of Pasadena, California, USA show that we can detect >70% of the street trees, assign correct species to >80% for 40 different species, and correctly detect and classify changes in >90% of the cases.

A novel downscaling approach to predict plant invasions and improve local conservation actions

To successfully protect native biodiversity from the effects of biological invasions, local conservation priorities must be established. For this purpose, fine-grained species distribution data is required but often unavailable. We present a new approach to obtain fine-grained predictions of invasion through the development of downscaled invasion maps based on coarse-grained distribution data. The framework is illustrated for the alien invader Acacia dealbata in the Northwest of Portugal. The analytical design was divided in five steps: (1) three individual coarse-grained models were calibrated and their spatial predictions were downscaled into fine-grained models using three different downscaling techniques; (2) a Downscaling Consensus Map was built by spatially combining the predictions from those three models; (3) using coarse-grained (1 km2) or fine-grained (0.04 km2) datasets, two different models were fitted and spatially projected; (4) for each spatial resolution, Conservation Value maps were produced, based on the spatial combination of the protection networks represented in the region; and (5) the spatial conflicts between the predicted distribution of the invader and Conservation Value maps were calculated and compared for the several invasion maps. The downscaled models showed high predictive performance (AUC > 0.9). The spatial projections of the different models revealed a general similarity among projections from all modelling techniques, for both the patterns of invasion and the conflicts with conservation areas. The possibility of obtaining detailed and reliable predictions based on coarse-grained distribution data could avoid costly fieldwork to collect fine-grained distribution data while effectively supporting the management of invasions at the appropriate scales.

Population biology of Elaeodendron transvaalense Jacq. in the presence of harvesting

Elaeodendron transvaalense Jacq. is one of the medicinal plant species commonly used by people in the Venda region. The species is known to treat a variety of diseases. It is widely used and important to traditional healers. Because of this, it found its way into the muthi markets, and it is among the seven most commonly traded plant species in the Venda region. This study investigated the impact of bark harvesting on the population structure of this species. The study revealed that although the level of bark harvesting is high, the species is able to resist that stress since it is a fine-grained species. The population also showed the ability to regenerate as it exhibited an inverse J-shaped curve. The crown health status was generally good although some individuals, contributing 9% of the sample, had dead crowns, which are a cause for concern. A linear relationship was noticed between areas harvested and stem circumference, which is understandable considering the large surface area of harvestable bark on bigger individuals.

Habitat selection by two sympatric rodent species in the Monte desert, Argentina. First data for Eligmodontia moreni and Octomys mimax

Habitat selection in deserts is mainly modulated by the availability of resources. The basic hypotheses regarding coexistence of species in arid environments offer two possibilities: 1) coexisting species partition available resources (each species acting as a specialist) and 2) generalists coexist with specialists because the former use habitats which are underused or not selected by the latter. This study presents the first data on macro- and microhabitat selection by Eligmodontia moreni (Cricetidae) and Octomys mimax (Octodontidae), two rodent species which coexist in the most arid part of the Monte desert of Argentina. In order to evaluate both hypotheses, animals were trapped during two seasons (humid and dry) in six macrohabitats of the arid Monte (creosote bush scrub, mesquite woodlands, columnar cactus slopes, barrens, chical and saltbush). A multivariate analysis of variance was used to evaluate differences in PCA components of microhabitat (floristic composition and vegetation structure) among capture and control points. The resulting models suggest E. moreni to be described as a fine-grained generalist species, which would be segregated from the specialist O. mimax in shared macrohabitats by resource partitioning, basically related to substrate type and vegetation cover. Our findings accord with theories of community assembly involving generalist and specialist species.

The effect of the spatial scale of recruitment on tree diversity in Afromontane forest fragments

In fragmented landscapes, tree recruitment is critical for forest persistence. We examined the effects of disturbance and environmental factors, isolation distance, and forest area on the spatial scale (grain) of regeneration of tree species in Afromontane forest fragments in South Africa. A species’ grain is defined by whether it typically regenerates within its own canopy shadow (fine-grained) or over a larger spatial scale (coarse-grained). Species richness did not differ between small and large forest fragments but there were proportionately more coarse-grained species and fewer fine-grained species in small than in large fragments. While coarse-grained species richness increased with decreasing disturbance and increasing fragment isolation, fine-grained species richness increased with increasing fragment area. Fine-grained species are vulnerable to area-dependent fragmentation pressures. Although they regenerate in their canopy shadow, fine-grained species do not dominate disturbed fragments as expected. While able to survive in small fragments, fine-grained species are potentially dispersal limited and are not good colonisers and depend for their persistence on establishment in large forests. Conversely, because coarse-grained species have effective dispersal mechanisms they can colonise small fragments and are important for the maintenance of tree diversity in fragmented Afromontane forests. Thus, consideration of species grain of regeneration is necessary in conserving Afromontane tree diversity. Fine-grained species are conserved by protecting large forests while coarse-grained species are effectively conserved by maintaining small forest fragments often assumed to be ecologically unviable.

Floristic heterogeneity of flooding Pampa grasslands: a multi-scale analysis

The present paper describes vegetation heterogeneity at contrasting spatial scales in order to analyse patterns of floristic diversity and vegetation – environment relations in the Flooding Pampa grasslands of Argentina. The overall grassland flora contains a large proportion of exotic species, which are mostly annual forbs. At a regional scale, there is a latitudinal turnover in relative abundance of major grass tribes, and C3 grasses become increasingly common in grasslands at higher latitudes. The principal gradients in overall vegetation composition were associated with landscape-level changes in topography and soil chemistry. Both native and exotic species richness increased with topographic elevation, and markedly decreased in flood-prone habitats with acidic or saline soils. In remnants of tussock grassland, the predominant physiognomy of the Pampas in pre-settlement times, the presence of the tall native grass Paspalum quadrifarium is a strong determinant of local community structure, reducing the percentage of exotic species. Heterogeneity analysis at the community scale revealed a fine-grained species packing that was in sharp contrast with spatial mosaics observed in other grasslands under livestock grazing. At this scale, spatial variation in floristic composition correlated with changes in various soil parameters, but not with subtle differences in topographic relief. The studied grasslands possess some unique features that contribute towards a better understanding of how current vegetation heterogeneity is controlled by historical, environmental and management factors.

Subsistence harvesting of pole-size understorey species from Ongoye Forest Reserve, South Africa: Species preference, harvest intensity, and social correlates

We investigate the effect of subsistence harvesting on the ecological status of the pole-size tree component of the understorey at the Ongoye Forest Reserve (OFR; 2611 ha), KwaZulu-Natal province, South Africa. Using generalised linear modelling (GLM) we examine the ecological and social correlates of species preference and harvest intensity. Data were collected from 22 strip transects (5 m × 300 m; 0.15 ha). Only 11.6% of the available pole-size trees (2 cm < DBH < 15 cm) were harvested, mostly for building materials. No instance of canopy tree logging was recorded. Sixty-eight species were identified; however, only seven species (82% of harvested stems) were preferred: Englerophytum natalense (33%), Garcinia gerrardii (19%), Drypetes gerrardii (9%), Tabernaemontana ventricosa (9%), Rinorea angustifolia (4%), Oxyanthus speciosus (4%), and Chrysophyllum viridifolium (4%). Size-class distributions for these seven species were inverse J-shaped, typical of fine-grained species that regenerate over small spatial scales, suggesting that current harvesting intensities may be sustainable. Pole-size stem density was significantly similar among residual stands in harvested areas (2014 ± 31 stems ha −1) suggestive of a stem-density harvest threshold below which further effort was unprofitable. The number of harvested stems increased with increasing stem availability across species and stem size-classes. Small size-classes (2–5 cm DBH) were harvested most intensely, followed by the intermediate (5–10 cm) and the large (10–15 cm) size-classes, for all species except C. viridifolium. For the latter, the harvesting intensity was greatest for the 10–15 cm size-class. Lastly, harvest intensity was greatest in those areas closest to households near the reserve boundary but was not affected by household density. Although subsistence harvesting at the OFR appears to be sustainable at current levels, we note that similar harvest intensities of pole-sized stems in studies from smaller forests (<60 ha) led to local extinction of tree species. In addition, because the dominant understorey species at OFR are almost exclusively harvested, the mid- to long-term effects of this harvesting preference on forest dynamics must be assessed to develop sound ecological forest management policies.

Using the spatial grain of regeneration to select harvestable tree species in subtropical forest

Abstract We describe a graphic model for selecting tree species suitable for harvesting of pole-sized stems. Pole size stems are the most harvested size class by rural users of African forest. Increasing harvesting pressure on this size class has created an urgent need for tree species selection and harvest off-take strategies that differ from large-scale and canopy tree-logging processes. Methods for selecting species suitable for harvesting are constrained in their complexity by the need to use available data that are typically from single survey, size-class frequency distributions (SCDs) of population structure. Using SCD data sets from three subtropical Afromontane forests in South Africa, we use the spatial scale (grain) of regeneration of the harvestable fraction of pole-sized stems to identify species that can be harvested. In fine-grained species the sub-canopy individuals are well represented beneath the canopy of mature conspecifics and thus display high levels of natural regeneration over a small spatial scale. Fine-grained species are suitable for harvesting because there is a large supply of pole-sized trees whose exploitation has potentially less effect on forest structure and composition than the use of coarse-grained species. Furthermore, because the abundant fine-grained species may experience strong neighbourhood effects in time, and their stem densities are likely to be regulated by density-dependent factors, judicious exploitation of these species is arguably an extension of the natural process of density-dependent self-thinning. In contrast, coarse-grained species regenerate over large areas and at low densities and their use requires careful management and should probably be restricted. Unfortunately, most species being targeted for harvesting are coarse- to medium-grained and appear to regenerate at levels insufficient for replacement. Using the spatial grain of regeneration to select harvestable species does not provide estimates of sustainable harvesting rates. These critical harvesting limits (stems ha−1 per annum) are set from experience. Incorporated into the management process, this model provides a simple means of ensuring that levels of natural regeneration are maintained without the need for intervention to maintain either tree species diversity, or potential economic value for pole production, in heavily used subtropical forests.

The dynamics and sustainable use of high-value tree species of the coastal Pondoland forests of the Eastern Cape Province, South Africa

We investigate the ecological status of the standing crop of high-value tree species in coastal scarp forest of the Eastern Cape Province of South Africa. We compare resource availability, distribution and use in several forests that are under state and community management. Species suitable for harvesting were identified using a simple linear-programming model based on the frequency and scale of regeneration (grain) of the tree species. Twenty high-value tree species were widely used. Nine species were apparently over-exploited and absent from community forests. Tree densities were higher in state forests than community forest. Size-class distribution profiles for all species, except Harpephyllum caffrum and Heywoodia lucens, displayed the characteristic inverse J-shaped distribution that generally indicates stable populations. Tree species varied in their spatial scale of recruitment, but were mostly intermediate- to coarse-grained. Two fine-grained species in state forest ( Englerophytum natalense and Millettia grandis) and only one species in community forest ( E. natalense) had densities that could potentially sustain current levels of exploitation. The prevailing tenurial system (i.e. state or community forest), the tree species and category of stem size (dbh) used, all affected tree harvesting off-take levels. Trees in the 10–20 cm diameter size-class (poles) were most harvested. Harvesting levels of pole-sized stems were more influenced by tree size than species, were significantly greater in community than state forests, and could be explained by the closer proximity of households to community ( x ̄ =1.5 km) than state forests ( x ̄ =3.2 km). Given the willingness of communities distant from forests to plant multipurpose indigenous trees, there is the potential for strengthening community forestry, and thus rural development, via forestry interventions such as agroforestry. However, the survival of these forests depends on the establishment of suitable integrated community-based institutions and management practices.

XAFS determination of the chemical form of lead in smelter-contaminated soils and mine tailings; importance of adsorption processes

We investigated smelter-contaminated soils from Evin-Malmaison, Nord-Pas-de-Calais, France, and mine tailings from Leadville, Colorado, U.S.A. Bulk Pb concentrations range from 460 to 1900 ppm in the topsoils at Evin-Malmaison site and from 6000 to 10000 ppm in the tailings samples from the Leadville site. These concentrations necessarily raise human health and environmental concerns, but bioavailability and chemical lability of Pb in these materials vary dramatically and show little correlation with bulk concentrations. This study provides detailed information on the speciation of Pb in these materials. Emphasis is on the identification and characterization of poorly crystalline and/or fine-grained species, such as sorption complexes and poorly crystalline (co)precipitates, which are likely to control Pb bioavailability and mobility in these natural systems. Because these samples are heterogeneous, multi-phase mixtures, a variety of bulk analytical methods were used including powder X-ray diffraction (XRD) and Rietveld refinement, scanning electron microscopy, electron probe microanalysis, synchrotron-based X-ray absorption, and micro-fluorescence spectroscopies. The synchrotron-based techniques enable identification of amorphous or nanocrystalline Pb-containing phases and the spatial distribution of Pb at the 25 mu m scale. These techniques, in conjunction with physical and chemical separation techniques, allowed identification and characterization of several species not amenable to detection by conventional microanalytical techniques. In the Evin-Malmaison samples, direct spectroscopic evidence for Pb sorbed to humic acids was found, as well as to both manganese and iron (oxyhydr)oxides. In the Leadville samples, variations in Pb speciation with pH are consistent with predictions based on simplified model system studies of adsorption processes: specifically, the carbonate-buffered tailings with near-neutral pH contain up to 50% of total Pb as adsorption complexes on iron (oxyhydr)oxides, whereas Pb speciation in sulfide-rich low pH samples is dominated by Pb-bearing jarosites with no evidence for adsorbed Pb in these latter samples.