AI for Forests: Monitoring & 3D Data

Artificial Intelligence (AI) is transforming the field of forestry by enabling more efficient and precise management of forest resources. AI-driven tools such as remote sensing technologies and Geographic Information Systems (GIS) allow for real-time monitoring and analysis of forest health, biodiversity, and biomass. The integration of 3D data in forestry is revolutionizing forest management by improving visualization, data accuracy, and stakeholder communication.

3D data is directly related to AI as it is often utilized in conjunction with AI algorithms to enhance various applications in forestry, such as improving the accuracy of tree detection, forest mapping, and biomass estimation through advanced data processing and analysis techniques.

Hence, today we'll have a closer look at studies related to 3D data in forestry used for monitoring natural forests, as well as a case study on the application of computer vision for tree detection in urban forests.

3D Data in Virtual Forests: A Comprehensive Review

Let's start with the review, recently published by a group of Swiss ???? scientists (with the contribution from University of Helsinki ????). Their study aimed to review the emerging questions and applications of 3D data in forestry through a comprehensive analysis of existing 3D technologies and their interactions with users and stakeholders.

The researchers have done a substantial work on a thorough review of state-of-the-art 3D reconstruction and visualization tools within the context of forestry. They were utilizing a preliminary analysis on research keywords to track trends and challenges in the field. The study assessed data acquisition methods, the application of digital 3D data, and its potential future interactions with specific demands from the forestry sector.

The key findings highlight that the use of digital 3D data in forestry is increasing, with significant potential for enhancing communication between stakeholders.

The review identified several emerging questions, including the challenges related to data acquisition and the integration of 3D data into forest management practices.

The study emphasized the need for improved methods in 3D data processing to better support forest visualization and management.

Practitioners in forestry, including researchers, forest managers, and policymakers, can apply the results of this research to enhance the efficiency and effectiveness of forest management and conservation efforts.

Accurate Tree Detection in Forests Using Virtual Datasets

The group of American ???? scientists has recently published results of their study which focused on development and evaluation of a multi-modal forest monitoring network (M2fNet) using a newly simulated large-scale virtual forest dataset to improve tree detection and segmentation tasks.

The first question: why virtual datasets? The answer is clear: virtual datasets provide high-fidelity, perfectly annotated data that enhance the training and performance of machine learning models for forest monitoring, overcoming the limitations of manual annotation and real-world data scarcity.

This virtual forest dataset has been created with 19 tree species using Unreal Engine, which generated 50,000 video frames with synchronized multimodal data (RGB, depth, and masks). The M2fNet model, leveraging hierarchical vision transformers (Swin Transformers), was trained on this dataset to extract and fuse RGB and depth features for instance-level tree detection and segmentation. Evaluation metrics, including mean average precision (mAP) at an IoU threshold of 50%, were used to assess model performance. Additionally, a user study with domain experts was conducted to validate the realism and utility of the simulated data for forestry applications.

This research successfully demonstrates that the M2fNet model significantly improves tree segmentation accuracy, achieving high mAP values and outperforming models trained solely on RGB imagery. The simulated dataset effectively mitigates overfitting issues common with limited real data, enhancing the reliability of forest monitoring tasks such as tree diameter measurement, long-term tree tracking, and forest mapping.

The user study confirmed the photorealistic quality of the simulated forest scenes, supporting their use in forestry education and research.

The results of this research can be practically applied by forestry researchers, environmental scientists, and AI developers to enhance precision forestry, improve forest management practices, and support sustainable forestry initiatives through advanced monitoring and analytical capabilities.

Urban Tree Detection in Street-View Images

The final research for today is related to urban forestry. Recently published study, conducted by Chilean ???? scientists, aimed to evaluate the application of deep learning techniques for automating the identification and classification of urban trees using street-view images.

The methodology involved using a combination of custom and publicly available datasets, including Arbocensus, Jekyll, and Barknet, which contain images of urban trees and their trunks. The research utilized ResNet-50 for tree genus classification and segmentation models such as DeepLabV3 and Segformer for pixel-wise segmentation of trees and trunks. The models were fine-tuned using transfer learning techniques, leveraging pre-trained models on large datasets like ImageNet, ADE20k, and Cityscapes, followed by training on the specific urban tree datasets. The study employed data augmentation strategies to enhance model robustness and utilized quantitative metrics like precision, recall, and Intersection over Union (IoU) to evaluate model performance.

The key findings indicate that the ResNet-50 model trained with the Arbocensus dataset achieved an average accuracy of over 0.90 in tree genus classification, with Platanus and Prunus being the easiest genera to identify and Liquidambar being the most challenging. Segmentation models achieved an IoU of up to 0.88 for tree segmentation, with the best results obtained using Segformer pre-trained with Cityscapes.

However, segmentation performance decreased in complex urban environments with occluded or overlapping trees. Trunk segmentation models showed that training on local datasets provided better results than transfer learning from external datasets, with an IoU of approximately 0.81 for the Arbocensus dataset.

These results can be practically applied by urban planners, arborists, and environmental researchers to improve urban tree management, monitor tree health, and enhance urban biodiversity assessments through automated tree detection and classification.

?? What's next in AI for Forests?

In the next 'AI for Forests' edition, we will continue to explore monitoring. We'll try to find some interesting research papers that demonstrate the successful application of remote sensing technologies in forestry

?What do you think about this topic?

Please, share your thoughts with us in the comments below ??

Wishes of insightful monitoring for your forests,

Maryna Kuzmenko, Ph.D ????, Chief Inspiration Officer at Petiole Pro Community

#forestry

Photo credit for the cover:

Arevalo-Ramirez, T., Alfaro, A., Figueroa, J., Ponce-Donoso, M., Saavedra, J. M., Recabarren, M., & Delpiano, J. (2024). Challenges for computer vision as a tool for screening urban trees through street-view images. Urban Forestry & Urban Greening, 95, 128316. https://doi.org/10.1016/j.ufug.2024.128316

AI in Forestry - References