Jacob Kamminga

Data shortages fuel controversy about an ongoing insect biodiversity crisis. Insects are immensely diverse and functionally critical for ecosystems, yet data on their trends remain patchy and biased. Sensors, ranging from camera traps and acoustic recorders to weather radar stations, are set to transform data collection in entomology. Meanwhile, AI models that extract biological information from sensors are improving at a startling rate. Realising the potential of automated monitoring means progressing from proof‐of‐concept studies to scalable insect sensing systems. However, stakeholders face severe operational challenges when adopting a growing suite of sensors, models and protocols for insect surveillance. Deployment of devices is not well co‐ordinated, while the risks of relying on AI are overlooked or understated. To achieve monitoring goals, common pitfalls related to sensors and AI need to be exposed and avoided. We trace a seven‐step path towards an effective transnational rollout of insect sensing systems. Step (1) reviews strengths, weaknesses and synergies across visual, acoustic, radar and photonic sensors; (2) confronts species determination—a key challenge for sensors and AI—suggesting how to improve identification and make use of uncertain data; (3) promotes creating and sharing standardised, labelled data, offering ecological insights and opportunities to train and evaluate AI. Focussing on data processing and scalability, step (4) proposes using AI to complement other resources, both human and digital, given it is not always the best tool for the job; (5) highlights how and why AI models fail, calling for routine evaluations and shrewdness during model training; (6) aims to break barriers to wider uptake of technologies through knowledge sharing, affordable design principles and equitable computing infrastructures, and finally, (7) emphasises that any revolution in insect monitoring must be grounded in good sampling design, with established monitoring schemes at its core. We set a trajectory for coordinated development of insect sensing systems, focussing not only on technical performance but also on integration with human expertise, case‐based evaluation and harmonisation with historical long‐term datasets. We address fundamental challenges of sensors and AI for biodiversity monitoring, producing recommendations that apply to all branches of the tree of life.

High-performing out-of-distribution (OOD) detection, both anomaly and novel class, is an important prerequisite for the practical use of classification models. In this paper, we focus on the species recognition task in images concerned with large databases, a large number of fine-grained hierarchical classes, severe class imbalance, and varying image quality. We propose a framework for combining individual OOD measures into one combined OOD (COOD) measure using a supervised model. The individual measures are several existing state-of-the-art measures and several novel OOD measures developed with novel class detection and hierarchical class structure in mind. COOD was extensively evaluated on three large-scale (500k+ images) biodiversity datasets in the context of anomaly and novel class detection. We show that COOD outperforms individual, including state-of-the-art, OOD measures by a large margin in terms of TPR@1% FPR in the majority of experiments, e.g., improving detecting ImageNet images (OOD) from 54.3% to 85.4% for the iNaturalist 2018 dataset. SHAP (feature contribution) analysis shows that different individual OOD measures are essential for various tasks, indicating that multiple OOD measures and combinations are needed to generalize. Additionally, we show that explicitly considering ID images that are incorrectly classified for the original (species) recognition task is important for constructing high-performing OOD detection methods and for practical applicability. The framework can easily be extended or adapted to other tasks and media modalities.

ARISE (Authoritative and Rapid Identification System for Essential biodiversity information) is a digital infrastructure with the mission to provide semi-automated identification of all multicellular species in the Netherlands and innovate on biodiversity monitoring solutions. By applying DNA barcoding, artificial intelligence for species recognition based on images, sound and radar, and emerging technology for biodiversity monitoring, we can radically speed up the inventory of life on earth. The infrastructure relies on a species reference database, which brings together new and existing biodiversity information on all multicellular species in the Netherlands. ARISE will be open-access and organize data according to FAIR principles (findable, accessible, interoperable and reusable). We will develop the infrastructure for the Dutch research community with financial support from the Dutch Research Council. ARISE is a partnership between the Naturalis Biodiversity Center, the University of Amsterdam, Westerdijk Institute and the University of Twente. With an Agile development approach, we will build iteratively guided by use cases and real-life challenges. Facilities and services include streamlined and accesible species sampling and (meta)barcoding, machine learning services to analyse digital input and a digital platform where all data and services come together powered by a state of the art data management system based on digital objects. We will be driven by continuous feedback from external scientists and representatives of species monitoring organizations. With ARISE, we will provide the scientific community with the foundation for automated recognition and monitoring of the entire Dutch biodiversity. Furthermore, ARISE will provide a basis for better biodiversity monitoring to provide policy makers with more reliable information on biodiversity and ecosystem change. This presentation will give an overview of work in progress, demonstrating several pilot versions of facilities and services, and what we envision to build in the future.

Movement data were collected at a riding stable over seven days. The dataset comprises data from 18 individual horses and ponies with 1.2 million 2-s data samples, of which 93,303 samples have been tagged with labels (labeled data). Data from 11 subjects were labeled. The data from six subjects and six activities were labeled more extensively. Data were collected during horse riding sessions and when the horses freely roamed the pasture over seven days. Sensor devices were attached to a collar that was positioned around the neck of horses. The orientation of the sensor devices was not strictly fixed. The sensors devices contained a three-axis accelerometer, gyroscope, and magnetometer and were sampled at 100 Hz.

Between 1960 and 1990, 95% of the black rhino population in the world was killed. In South Africa, a rhino was killed every 8 h for its horn throughout 2016. Wild animals, rhinos and elephants, in particular, are facing an ever increasing poaching crisis. In this paper, we review poaching detection technologies that aim to save endangered species from extinction. We present requirements for effective poacher detection and identify research challenges through the survey. We describe poaching detection technologies in four domains: perimeter based, ground based, aerial based, and animal tagging based technologies. Moreover, we discuss the different types of sensor technologies that are used in intruder detection systems such as: radar, magnetic, acoustic, optic, infrared and thermal, radio frequency, motion, seismic, chemical, and animal sentinels. The ultimate long-term solution for the poaching crisis is to remove the drivers of demand by educating people in demanding countries and raising awareness of the poaching crisis. Until prevention of poaching takes effect, there will be a continuous urgent need for new (combined) approaches that take up the research challenges and provide better protection against poaching in wildlife areas.

Audio latency, defined as the time duration when an audio signal travels from the microphone to an app or from an app to the speakers, significantly influences the performance of many mobile sensing applications including acoustic based localization and speech recognition. It is well known within the mobile app development community that audio latencies can be significant (up to hundreds of milliseconds) and vary from smartphone to smartphone and from time to time. Therefore, it is essential to study the causes and effects of the audio latency in smartphones. To the best of our knowledge, there exist mobile apps that can measure audio latency but not the corresponding status of smartphones such as available RAM, CPU loads, battery level, and number of files and folders. In this paper, we are the first to propose a framework that can simultaneously log both the audio latency and the status of smartphones. The proposed framework does not require time synchronization or firmware reprogramming and can run on a standalone device. Since the framework is designed to study the latency causality, the status of smartphones are deliberately and randomly varied as maximum as possible. To evaluate the framework, we present a case study with Android devices. We design and implement a latency app that simultaneously measures the latency and the status of smartphones. The preliminary results show that the latency values have large means (50 - 150 ms) and variances (4-40 ms). The effect of latency can be considerably reduced by just simply subtracting the offset. In order to achieve improved latency prediction that can cope with the variances an advanced regression model would be preferred.