@misc{StillerDuenasHempeletal., author = {Stiller, Stefan and Due{\~n}as, Juan F and Hempel, Stefan and Rillig, Matthias C. and Ryo, Masahiro}, title = {Deep learning image analysis for filamentous fungi taxonomic classification: dealing with small datasets with class imbalance and hierarchical grouping}, series = {Biology Methods and Protocols}, volume = {9}, journal = {Biology Methods and Protocols}, number = {1}, publisher = {Oxford University Press (OUP)}, issn = {2396-8923}, doi = {10.1093/biomethods/bpae063}, pages = {11}, abstract = {Deep learning applications in taxonomic classification for animals and plants from images have become popular, while those for microorganisms are still lagging behind. Our study investigated the potential of deep learning for the taxonomic classification of hundreds of filamentous fungi from colony images, which is typically a task that requires specialized knowledge. We isolated soil fungi, annotated their taxonomy using standard molecular barcode techniques, and took images of the fungal colonies grown in petri dishes (n = 606). We applied a convolutional neural network with multiple training approaches and model architectures to deal with some common issues in ecological datasets: small amounts of data, class imbalance, and hierarchically structured grouping. Model performance was overall low, mainly due to the relatively small dataset, class imbalance, and the high morphological plasticity exhibited by fungal colonies. However, our approach indicates that morphological features like color, patchiness, and colony extension rate could be used for the recognition of fungal colonies at higher taxonomic ranks (i.e. phylum, class, and order). Model explanation implies that image recognition characters appear at different positions within the colony (e.g. outer or inner hyphae) depending on the taxonomic resolution. Our study suggests the potential of deep learning applications for a better understanding of the taxonomy and ecology of filamentous fungi amenable to axenic culturing. Meanwhile, our study also highlights some technical challenges in deep learning image analysis in ecology, highlighting that the domain of applicability of these methods needs to be carefully considered.}, language = {en} } @misc{RyoSchillerStilleretal., author = {Ryo, Masahiro and Schiller, Josepha and Stiller, Stefan and Palacio, Juan Camilo Rivera and Mengsuwan, Konlavach and Safonova, Anastasiia and Wei, Yuqi}, title = {Deep learning for sustainable agriculture needs ecology and human involvement}, series = {Journal of Sustainable Agriculture and Environment}, volume = {2}, journal = {Journal of Sustainable Agriculture and Environment}, number = {1}, publisher = {Wiley}, issn = {2767-035X}, doi = {10.1002/sae2.12036}, pages = {40 -- 44}, abstract = {Deep learning is an emerging data analytic tool that can improve predictability, efficiency and sustainability in agriculture. With a bibliometric analysis of 156 articles, we show how deep learning methods have been applied in the context of sustainable agriculture. As a general publication trend, China and India are leading countries for publication, international collaboration is still minor. Deep learning has been popularly applied in the context of smart agriculture across scales for individual plant monitoring, field monitoring, field operation and robotics, predicting soil, water and climate conditions and landscape-level monitoring of land use and crop types. We identified that the potential of deep learning had been investigated mainly for predicting soil (abiotic), water, climate and vegetation dynamics, but ecological characteristics are critically understudied. We also highlight key themes that can be better addressed with deep learning for fostering sustainable agriculture: (i) including above- and belowground ecological dynamics such as ecosystem functioning and ecotone, (ii) evaluating agricultural impacts on other ecosystems and (iii) incorporating the knowledge and opinions of domain experts and stakeholders into artificial intelligence. We propose that deep learning needs to go beyond automatic data analysis by integrating ecological and human knowledge to foster sustainable agriculture.}, language = {en} } @misc{SafonovaGhazaryanStilleretal., author = {Safonova, Anastasiia and Ghazaryan, Gohar and Stiller, Stefan and Main-Knorn, Magdalena and Nendel, Claas and Ryo, Masahiro}, title = {Ten deep learning techniques to address small data problems with remote sensing}, series = {International Journal of Applied Earth Observation and Geoinformation}, volume = {125}, journal = {International Journal of Applied Earth Observation and Geoinformation}, publisher = {Elsevier BV}, issn = {1569-8432}, doi = {10.1016/j.jag.2023.103569}, pages = {17}, abstract = {Researchers and engineers have increasingly used Deep Learning (DL) for a variety of Remote Sensing (RS) tasks. However, data from local observations or via ground truth is often quite limited for training DL models, especially when these models represent key socio-environmental problems, such as the monitoring of extreme, destructive climate events, biodiversity, and sudden changes in ecosystem states. Such cases, also known as small data problems, pose significant methodological challenges. This review summarises these challenges in the RS domain and the possibility of using emerging DL techniques to overcome them. We show that the small data problem is a common challenge across disciplines and scales that results in poor model generalisability and transferability. We then introduce an overview of ten promising DL techniques: transfer learning, self-supervised learning, semi-supervised learning, few-shot learning, zero-shot learning, active learning, weakly supervised learning, multitask learning, process-aware learning, and ensemble learning; we also include a validation technique known as spatial k-fold cross validation. Our particular contribution was to develop a flowchart that helps DL users select which technique to use given by answering a few questions. We hope that our review article facilitate DL applications to tackle societally important environmental problems with limited reference data.}, language = {en} } @misc{StillerGrahmannGhazaryanetal., author = {Stiller, Stefan and Grahmann, Kathrin and Ghazaryan, Gohar and Ryo, Masahiro}, title = {Improving spatial transferability of deep learning models for small-field crop yield prediction}, series = {ISPRS Open Journal of Photogrammetry and Remote Sensing}, volume = {12}, journal = {ISPRS Open Journal of Photogrammetry and Remote Sensing}, publisher = {Elsevier BV}, issn = {2667-3932}, doi = {10.1016/j.ophoto.2024.100064}, pages = {11}, abstract = {Predicting crop yield using deep learning (DL) and remote sensing is a promising technique in agriculture. In smallholder agriculture (<2 ha), where 84\% of the farms operate globally, it is crucial to build a model that can be useful across several fields (high spatial transferability). However, enhancing spatial model transferability in a small-scale setting faces significant challenges, including spatial autocorrelation, heterogeneity and scale dependence of spatial dynamics, as well as the need to address limited data points. This study aimed to test the hypothesis that spatial cross validation (SCV) is a more suitable model validation practice than random cross validation (RCV) to enhance model transferability for spatial prediction in a small-scale farming setting. We compared the performances of DL models that predict crop yield for several settings including three crop types and two DL architectures based on RCV with and without overlapping samples and SCV. Notably, we conducted model performance tests on external, equally sized fields instead of the field used for training. We used high resolution RGB imagery taken with a drone as input. Our results show that the models using SCV outperformed those using RCV when the models were tested on external fields (on average r = 0.37 for SCV, r = 0.18 for RCV with overlap and r = 0.07 without), even though the models using SCV showed a substantially lower performance for cross validation (CV) than those using RCV (r with SCV and RCV w/o overlap = 0.73 and 0.98/0.73, respectively). The results suggest that RCV leads to over-optimism by overfitting the spatial structure and remembering image-specific information (so called memorization). Our study offers the first empirical evidence in agriculture that SCV is preferable to RCV in small field settings for making DL models more transferable.}, language = {en} } @misc{SchillerStillerRyo, author = {Schiller, Josepha and Stiller, Stefan and Ryo, Masahiro}, title = {Artificial intelligence in environmental and earth system sciences : explainability and trustworthiness}, series = {Artificial intelligence review : an international science and engineering journal}, volume = {58}, journal = {Artificial intelligence review : an international science and engineering journal}, number = {10}, publisher = {Springer Netherlands}, address = {Dordrecht}, issn = {1573-7462}, doi = {10.1007/s10462-025-11165-2}, pages = {1 -- 23}, abstract = {Explainable artificial intelligence (XAI) methods have recently emerged to gain insights into complex machine learning models. XAI can be promising for environmental and Earth system science because high-stakes decision-making for management and planning requires justification based on evidence and systems understanding. However, an overview of XAI applications and trust in AI in environmental and Earth system science is still missing. To close this gap, we reviewed 575 articles. XAI applications are popular in various domains, including ecology, engineering, geology, remote sensing, water resources, meteorology, atmospheric sciences, geochemistry, and geophysics. XAI applications focused primarily on understanding and predicting anthropogenic changes in geospatial patterns and impacts on human society and natural resources, especially biological species distributions, vegetation, air quality, transportation, and climate-water related topics, including risk and management. Among XAI methods, the SHAP and Shapley methods were the most popular (135 articles), followed by feature importance (27), partial dependence plots (22), LIME (21), and saliency maps (15). Although XAI methods are often argued to increase trust in model predictions, only seven studies (1.2\%) addressed trustworthiness as a core research objective. This gap is critical because understanding the relationship between explainability and trust is lacking. While XAI applications continue to grow, they do not necessarily enhance trust. Hence, more studies on how to strengthen trust in AI applications are critically needed. Finally, this review underlines the recommendation of developing a "human-centered" XAI framework that incorporates the distinct views and needs of multiple stakeholder groups to enable trustworthy decision-making.}, language = {en} } @misc{SafonovaStillerYordanovetal., author = {Safonova, Anastasiia and Stiller, Stefan and Yordanov, Momchil and Ryo, Masahiro}, title = {Self-supervised learning outperforms supervised learning for crop classification by annotating only 5\% of images}, series = {Precision agriculture}, volume = {27}, journal = {Precision agriculture}, number = {1}, publisher = {Springer Science and Business Media LLC}, address = {Dordrecht}, issn = {1385-2256}, doi = {10.1007/s11119-025-10302-9}, pages = {1 -- 31}, abstract = {Purpose One of the most pervasive Artificial Intelligence (AI) methodologies utilized in the domain of agriculture for image-based classification purposes is Supervised Learning (SL). However, SL depends on a large amount of annotation effort and is susceptible to overfitting to the given prediction task. Self-Supervised Learning (SSL) is a novel training paradigm with the potential to address these issues, while its potential has not been investigated in the agriculture domain. This paper presents the initial experimental investigation and comparison of SL and SSL for the classification of agricultural images in the context of limited samples. Methods We used an agricultural subset of the Land Use and Cover Area Frame Survey (LUCAS) dataset serving as a case study. In total, it comprised 1,000 images for each of the 10 crops: common wheat, barley, oats, maize, potatoes, sugar beet, sunflower, rape and turnip rape, soya, and temporary grassland. For SL, we trained popular and frequently used Convolutional Neural Network (CNN) architectures such as VGG16, Inception, ResNet-18/50, SqueezeNet, ResNeXt-50, MobileNet-V2, ShuffleNet, EfficientNet-V2, and ConvNeXt Tiny with and without data augmentations. For SSL, the best-performing CNN architectures (ResNet-18, ResNet-50, and ResNeXt-50) were further tested. The architectures were pre-trained with the VICReg algorithm (Variance Invariance Covariance Regularization) and fine-tuned successively using supervision for crop type classification. Results Our results demonstrate that the SSL models can distinguish crop types (common wheat, barley, oats, maize, potatoes, sugar beet, sunflower, rape, soya, and grassland) even without labels based solely on morphological features and organize them into three semantically meaningful visual groups: cereal-like and grassland crops, upright broadleaf crops, and low-growing broadleaf crops. The fine-tuned models, particularly ResNeXt-50, achieved superior performance compared to any of the SLs. Notably, we show that the fine-tuned SSL models outperformed the best-performing SL models by using only 5\% of the labeled training data for fine-tuning, corresponding to a small and balanced subset of the training split. Conclusion These findings highlight the potential of SSL for improving classification efficiency and generalization under limited data availability conditions in agriculture applications, providing a viable path toward more efficient agricultural monitoring systems.}, language = {en} }