Article
Ecology
Ismail Kirbas, Ahmet Cifci
Summary: This study investigates the classification of wood species using convolutional neural networks and evaluates the performance of various deep learning architectures. The experimental findings demonstrate that the Xception model achieves remarkable performance on the WOOD-AUTH dataset.
ECOLOGICAL INFORMATICS
(2022)
Article
Plant Sciences
Geovanni Figueroa-Mata, Erick Mata-Montero, Juan Carlos Valverde-Otarola, Dagoberto Arias-Aguilar, Nelson Zamora-Villalobos
Summary: Tree species identification is vital for conservation, sustainable management, and combating illegal logging. This research developed a CNN-based automated tree species identification system with high accuracy. Additionally, an Android application was developed to identify tree species from images of wood cross-sections.
FRONTIERS IN PLANT SCIENCE
(2022)
Article
Environmental Sciences
Jixia Huang, Xiao Lu, Liyuan Chen, Hong Sun, Shaohua Wang, Guofei Fang
Summary: This paper proposes the use of deep convolutional neural networks (D-CNNs) for the detection of pine wood nematode disease in remote sensing images. Five popular models are trained using transfer learning on a dataset constructed from Gaofen-1 (GF-1) and Gaofen-2 (GF-2) images. The study shows that SqueezeNet performs better than other models in terms of transfer learning on the sample dataset. The model's recognition efficiency and accuracy are improved by referring to the Slim module structure, resulting in an improved model that can accurately and efficiently monitor pine wood nematode disease using remote sensing.
Article
Computer Science, Artificial Intelligence
Xin Jin, Yide Di, Qian Jiang, Xing Chu, Qing Duan, Shaowen Yao, Wei Zhou
Summary: The article proposes a method using a deep convolutional auto-encoder with special multi-skip connections in the YUV color space for image colorization. Experimental results show the effectiveness of the proposed method on different image datasets.
Article
Engineering, Electrical & Electronic
Jiao Wang, Chunrui Tang, Hao Huang, Hong Wang, Jianqing Li
Summary: Blind identification of channel codes is crucial in signal interception and intelligent communication systems. This paper proposes a deep residual network-based deep learning approach that achieves high recognition accuracy for various forms of convolutional codes without prior information. Experimental results demonstrate that this method outperforms traditional algorithms and existing DL-based algorithms in blind identification of convolutional codes.
DIGITAL SIGNAL PROCESSING
(2021)
Article
Zoology
Catarina Pinho, Antigoni Kaliontzopoulou, Carlos A. Ferreira, Joao Gama
Summary: Automated image classification is a thriving field in machine learning, with successful applications in dealing with biological images. This study focuses on the ability of these methods to identify morphologically similar species that are difficult to distinguish by humans. Deep-learning models are found to be highly accurate in field identification and monitoring of cryptic species complexes, reducing the workload of expert or genetic identification.
ZOOLOGICAL JOURNAL OF THE LINNEAN SOCIETY
(2023)
Review
Forestry
Jose Luis Silva, Rui Bordalo, Jose Pissarra, Paloma de Palacios
Summary: Wood identification is an important tool in various fields. Computer vision-based technology provides a fast and accurate method for wood identification, but its application in fields like cultural heritage is still limited.
Article
Spectroscopy
Jian-E Dong, Ji Zhang, Zhi-Tian Zuo, Yuan-Zhong Wang
Summary: A new method using two-dimensional correlation spectroscopy (2DCOS) and deep learning for the species discrimination of bolete mushrooms was proposed in this study. The method achieved high accuracy in identifying different species and a smartphone application was developed for practical use.
SPECTROCHIMICA ACTA PART A-MOLECULAR AND BIOMOLECULAR SPECTROSCOPY
(2021)
Article
Environmental Sciences
Janne Mayra, Sarita Keski-Saari, Sonja Kivinen, Topi Tanhuanpaa, Pekka Hurskainen, Peter Kullberg, Laura Poikolainen, Arto Viinikka, Sakari Tuominen, Timo Kumpula, Petteri Vihervaara
Summary: Over the past two decades, forest monitoring and inventory systems have shifted from field surveys to remote sensing-based methods. Current analysis methods have limitations, with deep learning methods showing greater potential. This study focused on the classification of major tree species and achieved high accuracy using 3D convolutional neural networks in hyperspectral data.
REMOTE SENSING OF ENVIRONMENT
(2021)
Article
Biology
Yantong Chen, Xianzhong Zhang, Zekun Chen, Mingzhu Song, Junsheng Wang
Summary: This paper proposes a fine-grained classification method for fly species in complex natural environments based on deep convolutional neural networks. By locating fly positions, extracting features, and integrating global and local information, the method achieves precise classification. Experimental results indicate that the proposed method has high accuracy in classifying fly species.
COMPUTERS IN BIOLOGY AND MEDICINE
(2021)
Article
Mathematics, Interdisciplinary Applications
Arshpreet Kaur, Vinod Puri, Kumar Shashvat, Ashwani Kumar Maurya
Summary: The objective of this study is to automate the identification of inter-ictal activity in patients with epilepsy and distinguish it from the activity of controlled patients and patients with artifacts. The developed Residual neural network architecture achieved outstanding results in identifying inter-ictal activity using scalogram inputs.
CHAOS SOLITONS & FRACTALS
(2022)
Article
Forestry
Zhongmou Fan, Wenxuan Zhang, Ruiyang Zhang, Jinhuang Wei, Zhanyong Wang, Yunkai Ruan
Summary: This study addresses the disorderliness issue in tree species classification using point cloud data by transforming it into projected images. The study investigates the influence of incorporating multiple distinct projection perspectives, integrating depth information, and utilizing various classification models on the classification of tree point cloud projected images. The results show that utilizing datasets with depth and multidirectional information can enhance the accuracy and robustness of image classification.
Review
Biochemical Research Methods
Sung-Wook Hwang, Junji Sugiyama
Summary: The advancements in computer vision and machine learning have revolutionized scientific disciplines and created a new research field in wood science known as computer vision-based wood identification. Research has reviewed mainstream studies using machine learning procedures to familiarize wood scientists with this area and help them choose appropriate techniques in wood science.
Article
Computer Science, Information Systems
Poornima Singh Thakur, Tanuja Sheorey, Aparajita Ojha
Summary: This paper introduces a lightweight Convolutional Neural Network (CNN) for automatic plant disease identification. The proposed model outperforms other deep learning approaches on multiple public datasets.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Computer Science, Information Systems
Xuelei Li, Rengang Li, Yaqian Zhao, Jian Zhao
Summary: This study reveals that reducing the relevancy between filters in the same layer of residual convolutional neural networks can improve image recognition accuracy. The proposed training method controls the update of filter weights, leading to better generalization ability and higher recognition accuracy.
MULTIMEDIA TOOLS AND APPLICATIONS
(2021)
Article
Forestry
Malgorzata Danek, Monika Chuchro, Tomasz Danek
Summary: The study shows that larches in the Sudetes are more vulnerable to climate changes, mainly due to droughts that are likely to intensify in the future. Comparisons of extreme growth reactions in larch from the Carpathians and the Sudetes reveal similarities in drivers and predicted climate changes, indicating that negative extreme responses will also be observed in the Carpathians in the near future. The results suggest that the growth of larch stands in both regions will be negatively affected by predicted climate changes.
TREES-STRUCTURE AND FUNCTION
(2021)
Article
Engineering, Biomedical
Adrian Kucharski, Anna Fabijanska
Summary: The study proposes a fully automatic pipeline combining the watershed algorithm and an encoder-decoder convolutional neural network for corneal endothelial cell segmentation. It achieves promising results on a heterogeneous dataset, outperforming the competitor in terms of cell detection accuracy and other metrics.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2021)
Article
Computer Science, Interdisciplinary Applications
Maciej Czepita, Anna Fabijanska
Summary: An automated pipeline was proposed for analyzing blood flow through retinal vessels to replace time-consuming manual methods. Using convolutional neural networks and full width at half maximum analysis, blood flow was successfully detected in 18 retinal blood vessels. The average difference between manual and automatic measurements was 4.96%, with an average relative error of 8% for single vessel measurements.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2021)
Article
Chemistry, Multidisciplinary
Abir Affane, Adrian Kucharski, Paul Chapuis, Samuel Freydier, Marie-Ange Lebre, Antoine Vacavant, Anna Fabijanska
Summary: Accurate liver vessel segmentation is crucial for hepatic diseases, with recent methods using deep learning like U-Net. This study compares 3D U-Net, Dense U-Net, and MultiRes U-Net on the IRCAD dataset, finding that full 3D processing is most accurate overall but slab-based MultiRes U-Net performs best among specific models.
APPLIED SCIENCES-BASEL
(2021)
Article
Environmental Sciences
Mateusz Zareba, Tomasz Danek
Summary: The historical analysis shows that pro-clean-air legislation and grassroots movement have a positive impact on air quality in Krakow, but smog still occurs in the city during late autumn, winter, and early spring. The study reveals that PM10 concentrations exceeded EU norms on several days in March 2021, with 50% of the carbon fraction in PM10 originating from domestic heating in Krakow. The migration of air pollutants from neighboring municipalities, where fossil fuel heating is allowed, contributes to this issue. The research utilized a low-cost sensor network to analyze PM10 concentrations and examine the relationship between PM10 concentrations and atmospheric characteristics.
AEROSOL AND AIR QUALITY RESEARCH
(2022)
Article
Geochemistry & Geophysics
Tomasz Danek, Bartosz Gierlach, Ayiaz Kaderali, Michael A. Slawinski, Theodore Stanoev
Summary: This paper presents a strategy for parameter selection in a multilayered model by comparing walkaway vertical seismic profiling data using the Bayesian information criterion. The study focuses on P-wave traveltimes and assumes elliptical polar velocity dependence. While a one-layer model with elliptical anisotropy yields good results, a more efficient tool for multilayer modeling is needed to improve inversion results. The article proposes two optimization steps to obtain the proper set of velocity values for specific parameterizations, by finding the signal trajectory and minimizing the misfit between the model and data. The best model is chosen based on the Bayesian information criterion.
GEOPHYSICAL PROSPECTING
(2023)
Article
Forestry
Malgorzata Danek, Tomasz Danek
Summary: Recent studies on the climate-growth relationship of larch in the Polish Sudetes indicate future growth limitations due to the increasing sensitivity to changing climatic factors, especially drought. The analysis shows a trend towards a more unified response of larch to dominant climatic factors over time, with increasingly positive impact of May temperature. The observed changes are linked to the rapid rise in temperature, which negatively affects water availability and is expected to restrict larch growth in the future.
TREES-STRUCTURE AND FUNCTION
(2022)
Article
Energy & Fuels
Mateusz Zareba, Tomasz Danek, Michal Stefaniuk
Summary: This paper presents a detailed analysis of the use of unsupervised machine learning techniques for reservoir interpretation based on geophysical measurements. Four clustering algorithms were compared using measurements with different vertical resolutions, and the results were validated through lithological identification of the medium based on drill core analysis.
Article
Multidisciplinary Sciences
Anna Fabijanska, Gabriel D. Cahalan
Summary: This study proposes a new fully automatic pipeline that quantifies the properties of resin ducts and tree-ring boundaries in Pinus trees. A convolutional neural network is used to detect the ducts and boundaries, and a region merging procedure is applied to identify connected components corresponding to successive rings. The proposed method achieves high detection sensitivity and precision for both resin ducts and tree-ring boundaries.
SCIENTIFIC REPORTS
(2023)
Article
Environmental Sciences
Mateusz Zareba, Hubert Dlugosz, Tomasz Danek, Elzbieta Weglinska
Summary: The study applies unsupervised machine learning algorithms to analyze spatiotemporal patterns of air pollution using big data collected from sensors in Krakow. The results reveal distinct differences between average and maximum values of pollutant concentrations. The study highlights the potential of machine learning techniques and big data analysis for identifying hot-spots, coldspots, and patterns of air pollution and informing policy decisions.
Article
Engineering, Biomedical
Adrian Kucharski, Anna Fabijanska
Summary: Currently, corneal endothelial image segmentation relies on convolutional neural networks, but the scarcity of labeled corneal endothelial data due to expensive cell delineation process limits their potential. This study proposes a method of synthesizing cell edges and corresponding images using generative adversarial neural networks, which has not been reported before. Experimental results on three datasets show that our solution provides a cost-effective and diverse source of training data for corneal endothelial image segmentation.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2023)
Article
Computer Science, Information Systems
Anna Fabijanska
Summary: The problem of image segmentation is crucial in computer vision, and deep-learning methods have become dominant in providing solutions. However, they require a large amount of costly training data. This paper proposes a semi-supervised image segmentation method using graph convolutional networks, which achieved good performance in binary and multi-label segmentation tasks.
Article
Multidisciplinary Sciences
Tomasz Danek, Elzbieta Weglinska, Mateusz Zareba
Summary: Despite restrictive laws, Krakow has the highest level of air pollution in Europe, with pollutants transported from neighboring municipalities. This study applied a complex geostatistical approach to analyze particulate matter concentrations. The results show the relationship between topography, meteorological variables, and PM concentrations, with wind speed and terrain elevation being the main factors. The study also examined pollution migration and sources through the analysis of the PM2.5/PM10 ratio.
SCIENTIFIC REPORTS
(2022)
