Arianna Salili-James
Senior Researcher, Machine Learning
Research Interests
Make your next move.
Prepare your application
Review drafts before you send them.
Explore related searches
Articles (14)
The Natural History Museum, UK (NHM), is at the forefront of digitizing vast natural history collections, with over six million of its 80 million specimens already digitized. Extensive, high‐quality, digital specimen datasets are crucial for the integration, and analysis of biological information, providing global accessibility and digital preservation. However, at current rates, it could take centuries to digitize entire collections. To accelerate this, researchers at NHM are exploring the use of collaborative robots (cobots) for digitization. Here, the focus is on the development of artificial intelligence (AI) pipelines for the digitization of one of the largest NHM collections: pinned insects. Aa proof‐of‐concept workflow is presented that leverages AI to assist in precise identification, handling, and digitization of insect specimens and labels. The pipeline is designed to be adaptable across different museum specimen datasets, and to one day integrate seamlessly with the newly introduced cobot at NHM. Experimental results achieved accuracies of 0.95 for specimen identification, 0.79 for pinheads, and 0.92 for specimen labels, in independent image and video test sets. These results demonstrate the potential of this workflow in accelerating digitization efforts whilst prototyping novel cobot‐integrated digitization systems and advancing the biodiversity informatics for data creation and accessibility.
Year:
2026
The digitisation of natural history collections at scale raises a range of logistical and curatorial challenges. One concern is the physical expansion of storage infrastructure. As part of the Distributed System of Scientific Collections UK (Smith et al. 2022), the Natural History Museum London (NHM) will rehouse a significant proportion of its 36 million pinned entomological specimens (~130,000 drawers). Digitisation involves imaging every specimen and attaching a physical, unique, human- and machine-readable identifier (a small card label with a Data Matrix barcode). These barcodes must be readable from above, and their addition during digitisation can increase drawer occupancy if they exceed the existing footprint. Manual estimation of the footprint increase is impractical at this scale, given the range of circumstances associated with different specimens and drawers. To address this, the NHM has developed an AI-driven approach to estimate specimen drawer expansion and support resource planning. The deep learning pipeline can automatically detect drawer objects, referred to here as classes, including specimens, labels, barcodes, notes, unit trays, and drawers, from high-resolution images. Our dataset includes 11,090 digitised pinned Coleoptera drawers from the Index Lot collection (Natural History Museum 2014), representative of typical entomological collections. The AI pipeline supports both "pre-" and "post-" digitisation analysis by calculating the specimen bounding areas, handling overlaps, and estimating the net footprint change. Our object detection models achieved high performance under real-world conditions, with a mean average precision (mAP*1) of 85.45% across all classes. Barcode detection reached 86.63% mAP, while the standard unit tray detection and unit tray type classification model achieved 99.5% accuracy. (An example of the detection outputs is shown in Fig. 1) Three different calculation methods were tested and evaluated to estimate the drawer expansion area, corresponding to three specimen expansion rates for specimens with Data Matrix barcodes attached. Rate 1 (Per-Drawer Area via Polygon Masks*2): calculates expansion rates per drawer by measuring the total area occupied by specimens and their labels, ensuring overlaps are counted only once using polygon masks, as shown in Fig. 2. Rate 2 (Total-Based Expansion Rate): calculates expansion rates by averaging the occupancy areas of barcode-attached specimens across the entire dataset. Rate 3 (Per-Specimen Expansion Rate): averages the expansion rate per specimen using only bounding box*3 areas. Every specimen contributes equally, regardless of its size, to the final value. Rate 1 (Per-Drawer Area via Polygon Masks*2): calculates expansion rates per drawer by measuring the total area occupied by specimens and their labels, ensuring overlaps are counted only once using polygon masks, as shown in Fig. 2. Rate 2 (Total-Based Expansion Rate): calculates expansion rates by averaging the occupancy areas of barcode-attached specimens across the entire dataset. Rate 3 (Per-Specimen Expansion Rate): averages the expansion rate per specimen using only bounding box*3 areas. Every specimen contributes equally, regardless of its size, to the final value. Table 1 compares different expansion rates. Both per-drawer Rate 1 and per-specimen Rate 3 have confidence intervals. Fig. 3 and Fig. 4 show their distributions: Rate 1 is right-skewed, indicating modest area increases from overlap, while Rate 3 is bimodal*4, with a small peak near 0 (large specimens unchanged) and another around 0.3 (barcode additions on smaller specimens). The global total-based Rate 2 is the mean of the ratios without confidence intervals, providing a macro-level view. In addition to estimating drawer expansion and supporting budget planning, this tool has been designed in a modular fashion, allowing individual components to be used independently at different stages of the digitisation and curation w
Year:
2026
Introduction The wildlife trade significantly contributes to the ongoing biodiversity crisis. While charismatic taxa receive more attention (Colléony et al. 2017), reptiles, which are the most traded fauna, are often overlooked, despite the fact that approximately 90% of traded reptile species are captured from the wild (Marshall et al. 2020). Curtailing this trade begins with species identification, yet enforcement officers at border control facilities often lack the access to herpetological expertise, hindering law enforcement efforts. However, with recent advances in artificial intelligence (AI), particularly in computer vision, various automated species identification applications have emerged (Zhang et al. 2024). Thus far, computer vision models have successfully classified different reptile species. For example, Patel et al. (2020) used Fast R-CNN (Region-based Convolutional Neural Networks) and ResNet models to classify nine endemic snake species of the Galapagos Islands from live snake images to achieve an accuracy of 75%. However, existing research focuses on classifying images of live reptiles and has not ventured into classifying processed reptillian goods, where common distinguishing features such as colour, body shape and positioning are not present, posing significant challenges. Unlike previous studies, our aim is to classify species solely through the dorsal textural pattern of their skin, allowing classification without taxonomically diagnostic features. In this study, we focus on snakes and introduce a multi-model pipeline to identify illegal snake goods to the species level. Our approach involves training segmentation (identification of snake bodies within images) and classification (identification of distinct snake species) models to classify species based on skin texture patterns captured from live snake imagery. With this pattern-based classification approach, our trained model is generalised to recognise these same patterns in processed goods. This ensures that snake goods that have been processed artificially can still be classified despite losing features used to identify live reptile specimens. Methods To reliably build a model to classify snake goods, a large dataset of snake skins is required. However, we utilise live snake images, rather than illegal snake goods, as they are more easily accessible. Our pipeline includes four key stages to ensure the model learns skin textural patterns rather than factors like body shape or colour: isolating snakes and snake goods from their background through two segmentation models; production of the box image dataset (splitting segmented images into boxes) to ensure only skin pattern is used for model training (Fig. 1); training an EfficientnetB0 model (Tan and Le 2019) to classify snakes from the skin pattern of live snake data and application of the EfficientnetB0 model on snake goods. isolating snakes and snake goods from their background through two segmentation models; production of the box image dataset (splitting segmented images into boxes) to ensure only skin pattern is used for model training (Fig. 1); training an EfficientnetB0 model (Tan and Le 2019) to classify snakes from the skin pattern of live snake data and application of the EfficientnetB0 model on snake goods. The pipeline evaluates the model's performance across eight snake classes in two stages, when: fed live snake image data, and fed illegal snake goods image data fed live snake image data, and fed illegal snake goods image data Evaluation across these two points in the pipeline allows us to understand whether the model maintains its classification abilities from live snake images to illegal snake goods. Results and Discussion When testing the model’s classification ability when fed live snake image data, an overall accuracy of 0.77 and AUC-ROC (Area Under the Curve-Receiver Operating statistic) scores ranging from 0.86 to 0.95 were obtained. These scores indicate that although final class
Year:
2026
It is estimated that there are nearly 400 million herbarium specimens held across approximately 3,500 herbaria worldwide (Davis 2023). Over the past decade, many institutions have embarked on large-scale digitisation initiatives to robustly and efficiently digitise herbarium sheet collections. While these processes can be highly productive, reaching daily rates of 600–1,000 herbarium sheets in some programmes, there remains a risk and cost associated with re-digitisation, often caused by complications during the imaging stage. To address these challenges, the Natural History Museum, London (NHM) partnered with the Royal Botanic Garden Edinburgh (RBGE) to develop an Artificial Intelligence (AI)-based workflow aimed at providing enhanced quality control of herbarium sheet digitisation. This initiative was supported by the UK Government’s Department for Culture, Media and Sport (DCMS) through the NHM–DCMS AI Pilot Programme (Poon et al. 2026). The aim of this project was to develop an AI-assisted quality control framework (Zhang et al. 2025). Trained on collections at RBGE, the tool enables automated quality checks of digitised herbarium sheets. This is done by detecting missing objects (such as colour reference cards or barcodes), identifying cropping errors, and highlighting focus issues (Fig. 1). The pipeline integrates several computational techniques, including object detection, decision trees, classical statistics, and deep learning methods. In addition to post-digitisation analysis, a user-friendly interface was also developed to support in situ quality checks during the imaging stages. Model evaluation on an independent test set demonstrated the high accuracy and robustness of the AI-assisted framework (Zhang et al. 2025). The object detection model achieved an overall precision of 0.98, recall of 0.98, and mean Average Precision scores of 0.99 (mAP50) and 0.94 (mAP50–95) across all object categories. Key reference elements, including colour reference cards, barcodes, and rulers, were detected with near-perfect accuracy, ensuring precise object localisation. We note that as model development was done on RBGE collections alone, this meant that reference elements were similar across many herbarium sheets in the sample. Lastly, the issue detection models also performed strongly, achieving a precision of approximately 1.00 for out-of-focus detection, 0.98 for over-cropping, and 0.93 for under-cropping. The proposed workflow underwent further evaluation with herbarium digitisers at RBGE, which will inform refinements to the model, software interface, and the overall pipeline. The next step is to expand on this pilot project by testing the model with collections across different institutions, leading to more refinements for generalisability. This can also help us evaluate and estimate the total proportions of digitised herbarium sheets with issues. Finally, our global aim is to provide an AI-based quality control service that can be integrated into digitisation workflows not only at RBGE but across institutions worldwide.
Year:
2026
From November 2024 to March 2025, the Natural History Museum, London (NHM), supported by the UK Department for Culture, Media and Sport (DCMS), led a national programme to explore a new model for deploying artificial intelligence (AI) across the UK cultural and heritage sector. Inspired by the Museum’s in-house AI Lab Programme (Poon 2024), which has successfully applied AI to research, collections management, and public engagement within the NHM (e.g., Poon and Leong 2024, Salili-James 2024, He et al. 2024, Zhang et al. 2025c, Gallois et al. 2025), this national initiative extended that model to a broader institutional landscape. The programme examined how AI can support biodiversity research, collections management, accessibility, and public engagement, while addressing the practical realities of AI adoption, scalability, and long-term sustainability. In partnership with the Zoological Society of London, Lyme Regis Museum, National Museums Liverpool, the Royal Agricultural University, and the Royal Botanic Garden Edinburgh, the programme delivered five AI pilot projects. Each project developed an AI-driven solution to a critical challenge: automated species classification of illegally traded snake goods (Leong et al. 2025a); automatic data extraction from index cards (Leong et al. 2025a); AI-generated alt-text for digital collections (van Schalkwyk et al. 2025); a conversational chatbot trained on herbarium data (Zhang et al. 2025b); and a herbarium digitisation quality-control tool (Zhang et al. 2025a), respectively. These pilots demonstrated how AI can deliver both operational efficiencies and public benefit. Automated transcription and text extraction have streamlined digitisation efforts, reducing the time and resources required for archival documentation. AI-driven image processing has enhanced the quality control mechanisms for digitisation, ensuring higher standards in digital preservation. Computer Vision models have provided new capabilities in species classification and object identification, offering valuable insights for conservation and research. Additionally, AI-powered accessibility tools, such as automated alt-text generation and chatbot interaction, have improved inclusivity by making collections more accessible to diverse audiences. By engaging partners of different scales, locations, disciplines, and specialisations, the programme provided a rare opportunity to test AI in diverse institutional environments. This diversity enabled robust evaluation of prototypes and fostered knowledge exchange, skills development, and capacity building across the sector. Crucially, the collaborative model demonstrated the importance of co-design: tools developed in close dialogue with institutional priorities were more readily embedded in existing practices and had clearer pathways to sustainable integration. Beyond technical outcomes, the programme highlighted sector-wide considerations for responsible AI adoption. These included: the need for consistent data standards to support model training and interoperability; designing systems that augment, rather than replace, professional expertise; ensuring infrastructure and training are in place to support long-term use; and transitioning from short-term pilots to sustainable, maintainable solutions. Collectively, the five projects suggest how the sector might progress from isolated experimentation to a coordinated ecosystem of digital innovation. Although the pilots are at proof-of-concept stage, the projects have already demonstrated measurable improvements in workflows, data quality, and accessibility. They provide a framework that other organisations can adapt, offering practical entry points into AI adoption. At the same time, the work underscores the importance of continued investment in digital transformation to ensure that heritage institutions can keep pace with rapid technological change. In summary, the NHM-DCMS AI Pilot Programme has shown that AI can be responsibly
Year:
2026
The Anthropocene presents significant challenges for global biodiversity, public health and ecosystem stability. The wealth of publicly available near‐real‐time ecology and climate data can be used to monitor these challenges and allow practitioners to develop mitigation strategies. There is untapped potential to apply large language models (LLMs) to quantitative ecological and environmental datasets, enabling researchers and practitioners to use natural language queries to transform ecological observations into actionable insights for both conservation action and communication of results to diverse audiences. Advances in artificial intelligence (AI), and particularly in LLMs, offer emerging opportunities to address these challenges. LLMs are increasingly proficient at identifying patterns and semantic relationships within textual data and are highly customisable. Accessible AI tools can facilitate communication across research and policy sectors. Here, we present a roadmap for designing and implementing multi‐modal LLMs to answer ecological research questions. To build robust ‘virtual quantitative assistants’ capable of fast‐tracking data interpretation, we advocate for strategic planning, data stewardship practices, careful prompt engineering and model evaluation as key steps in the LLM development process. We discuss potential use‐case examples that apply the LangChain framework to analyse citizen science data. Using our LLM roadmap, we highlight the importance of iterative and strategic prompt engineering and agent selection, in addition to iteratively evaluating model output. As LLM software continues to evolve, its integration into ecological and environmental research can empower ecologists with purpose‐built tools that bridge the gap between data collection and actionable solutions.
Year:
2025
Synopsis Artificial intelligence (AI) is poised to revolutionize many aspects of science, including the study of evolutionary morphology. While classical AI methods such as principal component analysis and cluster analysis have been commonplace in the study of evolutionary morphology for decades, recent years have seen increasing application of deep learning to ecology and evolutionary biology. As digitized specimen databases become increasingly prevalent and openly available, AI is offering vast new potential to circumvent long-standing barriers to rapid, big data analysis of phenotypes. Here, we review the current state of AI methods available for the study of evolutionary morphology, which are most developed in the area of data acquisition and processing. We introduce the main available AI techniques, categorizing them into 3 stages based on their order of appearance: (1) machine learning, (2) deep learning, and (3) the most recent advancements in large-scale models and multimodal learning. Next, we present case studies of existing approaches using AI for evolutionary morphology, including image capture and segmentation, feature recognition, morphometrics, and phylogenetics. We then discuss the prospectus for near-term advances in specific areas of inquiry within this field, including the potential of new AI methods that have not yet been applied to the study of morphological evolution. In particular, we note key areas where AI remains underutilized and could be used to enhance studies of evolutionary morphology. This combination of current methods and potential developments has the capacity to transform the evolutionary analysis of the organismal phenotype into evolutionary phenomics, leading to an era of “big data” that aligns the study of phenotypes with genomics and other areas of bioinformatics.
Year:
2024
In the last decade, the Natural History Museum, UK (NHM), has been at the forefront of the digitisation of natural history collections, with almost six million of its 80 million specimens digitised. This momentous undertaking has led to numerous innovations on how to optimise digitisation workflows. One avenue that is currently being explored is the use of collaborative robots—cobots. Since acquiring a Techman TM5 900 robotic arm in 2023 (Scott et al. 2023), we have been experimenting with its capabilities. Experiments began with simple pick-and-place tasks, using artificial specimens. Next, we focused on two use-cases, based on the digitisation of shark teeth and pinned-insects. Both shark teeth and pinned insects are in abundance at the NHM, making their manual digitisation a tedious task. Currently, we have trained the cobot to pick up a specimen, move it elsewhere to photograph and scan the specimen, then move it back to its original position or to a new place. Thus far, this has all been coordinate-based. Focusing on pinned insect specimens, we have now begun training deep learning models to perform segmentation, classification, and tracking tasks on images and manually taken videos. Segmentation and classification tasks range from distinguishing specimens from one another within drawers, to classifying different pins, labels, and insects. Meanwhile, object tracking methods are utilised to track labels from videos taken around the specimen. By tracking different labels simultaneously from multiple frames, we can combine the views of the labels in order to obtain a full picture for each label (for example, using tools described in Salili-James et al. 2022). Thus far, our machine learning pipelines have proved successful, for example, with F1 scores of 96–98% to classify and segment insects and to locate pin heads from dorsal views. Soon, we will be establishing workflows that integrate computer vision (CV) and machine learning (ML) techniques directly with the robotic arm, with pipelines that could be applied to different datasets, and that can significantly enhance efficiency. Broadly, these pipelines can be split into four sections: Specimen Identification : CV/ML to locate individual specimens or certain parts of specimens e.g., pinheads within pinned insects. Handling : With custom grippers, the cobot can delicately pick up, move, and place specimens, to and from photography stations for high quality scanning. Imaging & Scanning : The cobot will scan and photograph specimens. As the camera moves around the specimen, ML is used to segment and track the specimen labels to image them from the optimal views. A built-in Optical Character Recognition process can also be integrated to perform automatic transcription from here. Identifiers : This step requires the cobot to attach identifier labels to specimens or drawers, after locating the optimal position to do this, using CV/ML. Specimen Identification : CV/ML to locate individual specimens or certain parts of specimens e.g., pinheads within pinned insects. Handling : With custom grippers, the cobot can delicately pick up, move, and place specimens, to and from photography stations for high quality scanning. Imaging & Scanning : The cobot will scan and photograph specimens. As the camera moves around the specimen, ML is used to segment and track the specimen labels to image them from the optimal views. A built-in Optical Character Recognition process can also be integrated to perform automatic transcription from here. Identifiers : This step requires the cobot to attach identifier labels to specimens or drawers, after locating the optimal position to do this, using CV/ML. In this talk, we will discuss the progress of the NHM’s cobot research and explore the future of robotics for the digitisation of natural history collections.
Year:
2024
Calcareous nannofossils serve as crucial indicators for establishing the biostratigraphic age of chalk macrofossil specimens in natural science collections. Better age control of specimens collected several hundred years ago enables us to uncover the dark data hidden within these collections and incorporate these data into current research projects, examining ecosystem response to past climate change. However, the manual identification of these microscopic organisms is laborious and subjective, and so we are harnessing deep learning techniques for automatic nannofossil detection and identification. This approach required the construction of a robust dataset, currently comprising over 100,000 labelled images, complemented by the development of multiple specialised deep learning models. While some models focus on detecting target species, others are dedicated to species classification. Evaluation on an independent test set showcases the efficacy of our methodology, with the current detection model achieving a balanced accuracy of 93%. Similarly, the classification model demonstrates robust performance, attaining an average balanced accuracy of 96%. Furthermore, as well as assisting with our biostratigraphic studies, the dataset of accurately labelled images has enabled us to test other aspects of ecosystem response. For example, examining morphometric changes in nannofossils over geological time can provide valuable insights into the potential impact of current global warming on modern phytoplankton assemblages (Mancini et al. 2021). This is particularly important for coccolithophores (Young et al. 2005), which play a critical role as primary producers in the global carbon cycle. A decrease in their size could lead to bottom-up ecosystem impacts and reduced carbon sequestration (Poulton et al. 2007, Krumhardt et al. 2017). Using our dataset, we conducted a deep-learning-enhanced automatic morphometric analysis focusing on the nannofossil species, Tranolithus orionatus . Our analysis revealed that two key morphometric parameters, minor axis and area size, showed statistically significant differences between the Cenomanian stage (approximately 100.5 to 93.9 million years ago) and the post-Cenomanian stages of the Late Cretaceous (approximately 93.9 to 66.0 million years ago). Kolmogorov-Smirnov tests (Massey 1951) between the two samples yielded p-values of 0.039 for the minor axis and 0.031 for the area size. Understanding these morphometric changes is crucial due to the close parallels between current climate projections and the warming and greenhouse climate of the Late Cretaceous, particularly the Cenomanian-Turonian boundary event (Arthur et al. 1990). Insights into how organisms changed morphologically during past periods of environmental stress can help us more effectively predict future responses of organisms under similar conditions (Razmjooei et al. 2020). These findings underscore the effectiveness of our approach in automating the identification and recognition of chalk nannofossils, helping to unlock natural science collections and to address key questions related to marine response to past climate change.
Year:
2024
The Natural History Museum in the UK (NHM) is home to more than 80 million objects spanning 4.5 billion years of history. Each of these contain a wealth of data, whether on specimen labels, index cards, registers and/or diaries. Transcribing and categorising this information can help unlock crucial research potential. To do this at scale, we turn to computer vision (CV) and Machine Learning (ML) techniques to automate this work. Over a million of the museum’s specimens are ornithological, including one of the largest and most comprehensive egg collections in the world. Representing 52% of known bird species, with over 300,000 clutches (where a clutch defines the total group of eggs laid in a nest), collected over the last 200 years, arguably make this the most important archive of avian environmental change data in existence(Norris et al. 2023). The eggs were historically catalogued using index cards, containing key information such as identification, collection date, locality and clutch size. A proportion of these egg cards have now been imaged and this led to the start of this project, focusing on a sample of 15,000 photographed egg cards (example seen in Fig. 1). Our initial approach used Google Vision to perform Optical Character Recognition (OCR) to transcribe all text with the egg cards. By focusing on textboxes around key terms (e.g., “Collector”), and using CV tools, we approximated boxes around every key category. Finally, each text segment was associated to a category box, followed by minor post-processing in order to extract (i.e., transcribe and categorise) the data. Here we successfully extracted the data within the sample, with a 98.6% average accuracy. Although our methods worked well for our sample, they did rely on consistency within the structures of cards. To expand the project further, and to mitigate the reliance on consistent structures within cards, we turned to Large Language Models (LLMs). This allowed us to explore automatic data extraction from different types of cards and labels, despite variation in the card structure, and even handle unknown categories of text. Consequently, the scope of the data collected was widened, such as adding ornithological specimen data (e.g., skins), as well as external datasets through collaboration with the British Trust for Ornithology, who manage the Nest Record Scheme (Crick et al. 2003), which holds decades of vital information on the progress of monitored nests in the UK. This index-card data-extraction project is just the beginning. As we expand our data extraction capabilities, our aim is to develop a novel pipeline that can be applied not just to avifauna-related cards, but any structured textual data, with the potential to unlock invaluable insights.
Year:
2024
Year:
2024
Year:
2024
Year:
2023
Year:
2023
Collaborators (5)
Emily C. Watt
University College London
Helen Hardy
Deputy Head, Digital, Data & Informatics
Natural History Museum
Sanson Poon
Senior Research Data Scientist
Natural History Museum
Agnese Lanzetti
University of Birmingham
Vincent Smith
Head of Digital, Data and Informatics
Natural History Museum
Make your next move.
Prepare your application
Review drafts before you send them.