Article
Ecology
Sanjay Kumar Gupta, Shivam Kumar Yadav, Sanjay Kumar Soni, Udai Shanker, Pradeep Kumar Singh
Summary: This study proposes an automated approach for multiclass weed identification using semantic segmentation to improve weed control techniques, reduce pesticide usage, and enhance crop yields. A novel multiclass weed dataset was created and four advanced deep learning models were evaluated, with the U-Net-based Inception-ReseNetV2 achieving the highest F1-score of 96.78%. These findings demonstrate the efficacy of the proposed approach in accurately identifying and categorizing weeds in agricultural fields.
ECOLOGICAL INFORMATICS
(2023)
Article
Agriculture, Multidisciplinary
Jiacai Liao, Ibrahim Babiker, Wen-fang Xie, Wei Li, Libo Cao
Summary: By combining background transfer learning and color-attention module methods, the dandelion segmentation method can effectively segment the dandelion with a satisfactory accuracy rate.
COMPUTERS AND ELECTRONICS IN AGRICULTURE
(2022)
Article
Engineering, Multidisciplinary
Seungbo Shim, Jin Kim, Gye-Chun Cho, Seong-Won Lee
Summary: This study proposes a deep learning-based crack detection technology that can better identify cracks in concrete structures. It also introduces a stereo-vision-based method to reconstruct the 3-dimensional shape of cracks.
STRUCTURAL HEALTH MONITORING-AN INTERNATIONAL JOURNAL
(2023)
Article
Agriculture, Multidisciplinary
Keenan Granland, Rhys Newbury, Zijue Chen, David Ting, Chao Chen
Summary: This paper presents an iterative training methodology, Automating-the-Loop, for semantic segmentation. It significantly reduces the labeling effort while achieving comparable performance to other methods.
COMPUTERS AND ELECTRONICS IN AGRICULTURE
(2022)
Article
Computer Science, Artificial Intelligence
John Brandon Graham-Knight, Corey Bond, Homayoun Najjaran, Yves Lucet, Patricia Lasserre
Summary: This paper proposes finding network architectures that achieve similar performance while promoting diversity for ensembling. Prediction performance and diversity of various network sizes and activation functions applied to semantic segmentation are explained. The study shows that performance and diversity can be predicted from neural network architecture using explainable boosting machines. A grid search of 144 models is performed, and many of the models exhibit no significant difference in mean performance within a 95% confidence interval. It is found that the best performing models have varied network architecture parameters and diversity between models can be achieved by varying network size. Moreover, homogeneous network sizes generally show positive correlation in predictions and larger models tend to converge to similar solutions.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Construction & Building Technology
Hyunjun Kim, Sung-Han Sim, Billie F. Spencer
Summary: This study introduces an advanced stereo vision framework using wide-angle and telephoto lenses for crack quantification and 3D reconstruction of concrete structures. A robust depth estimation strategy is also proposed and the performance was field validated.
AUTOMATION IN CONSTRUCTION
(2022)
Article
Plant Sciences
Natalia Sapoukhina, Tristan Boureau, David Rousseau
Summary: Despite the lack of annotated datasets of plant images with disease lesions, this study proposes a novel methodology for generating fluorescent images of diseased plants with automated lesion annotation. The U-Net model trained purely by a synthetically generated dataset efficiently segments disease lesions on fluorescent images of plant leaves, achieving a recall of 0.793% and an average precision of 0.723% on an empirical fluorescent test dataset. The use of synthetic data can facilitate the application of deep learning methods in precision crop protection and improve the generalization ability of deep learning models.
FRONTIERS IN PLANT SCIENCE
(2022)
Article
Chemistry, Analytical
Xiaogang Jia, Wei Chen, Zhengfa Liang, Xin Luo, Mingfei Wu, Chen Li, Yulin He, Yusong Tan, Libo Huang
Summary: Stereo matching is an important research field in computer vision. By integrating fast 2D stereo methods with accurate 3D networks, this study improves performance and reduces computational time effectively. The method strikes a balance between speed and accuracy, achieving significant improvement in accuracy while being faster than other existing stereo networks.
Article
Computer Science, Artificial Intelligence
Junran Peng, Qing Chang, Haoran Yin, Xingyuan Bu, Jiajun Sun, Lingxi Xie, Xiaopeng Zhang, Qi Tian, Zhaoxiang Zhang
Summary: Pre-training on large-scale datasets is increasingly important in computer vision and natural language processing. However, it is expensive to use large-scale pre-training for per-task requirements due to different application scenarios. This paper proposes GAIA-Universe (GAIA), a system that automatically generates customized solutions for perception tasks, based on heterogeneous downstream needs. It provides powerful pre-trained weights and searching models that conform to hardware constraints, computation constraints, specified data domains, and limited data points.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Interdisciplinary Applications
Dimitrios Psychogyios, Evangelos Mazomenos, Francisco Vasconcelos, Danail Stoyanov
Summary: This paper proposes a learning-based framework for jointly estimating disparity and binary tool segmentation masks. The experimental results show that supervising the segmentation task improves the network's disparity estimation accuracy. The domain adaptation scheme enables domain adaptation of the adjacent disparity task. The best overall multi-task model performs well on the test sets.
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2022)
Article
Chemistry, Analytical
Athanasios Anagnostis, Aristotelis C. Tagarakis, Dimitrios Kateris, Vasileios Moysiadis, Claus Gron Sorensen, Simon Pearson, Dionysis Bochtis
Summary: This study proposed an approach for orchard trees segmentation using a deep learning convolutional neural network variant, achieving automated detection and localization of tree canopies under various conditions. The best-trained model achieved 91%, 90%, and 87% accuracy for training, validation, and testing, and reached up to 99% performance levels in testing on never-before-seen orthomosaic images or orchards, demonstrating robustness.
Article
Chemistry, Analytical
Ricardo Cruz, Diana Teixeira e Silva, Tiago Goncalves, Diogo Carneiro, Jaime S. Cardoso
Summary: Semantic segmentation involves classifying each pixel based on a set of classes. Conventional models have equal focus on classifying easy and hard-to-segment pixels, which is inefficient. This study proposes a framework where the model first produces a rough segmentation and then refines patches of the image estimated as difficult to segment. The framework is evaluated on four datasets and four state-of-the-art architectures, showing a fourfold acceleration in inference time with some tradeoff in output quality.
Article
Environmental Sciences
Alexandru Pop, Victor Domsa, Levente Tamas
Summary: In this paper, a novel rotation normalization technique using an oriented bounding box for point cloud processing is proposed. It is used to create a point cloud annotation tool for part segmentation and trained on custom datasets for classification and part segmentation tasks. The method is successfully deployed on an embedded device with limited processing power and compared with other rotation-invariant features in noisy synthetic datasets. Our method offers more auxiliary information related to the object's dimension, position, and orientation while performing at a similar level.
Article
Chemistry, Multidisciplinary
Jun Zhou, Xing Bai, Qin Zhang
Summary: This paper proposes a method to address the issue of misclassification in semantic segmentation models. By calculating the relevance between different class pairs, the method infers the category of connected components and corrects the misclassifications made by deep learning models. Experimental results demonstrate the effectiveness of the method, improving the performance of various deep learning models.
APPLIED SCIENCES-BASEL
(2022)
Article
Green & Sustainable Science & Technology
Gary Storey, Qinggang Meng, Baihua Li
Summary: This paper presents a study on leaf and rust disease detection in apple orchards using Mask R-CNN, and highlights that a Mask R-CNN model with a ResNet-50 backbone provides good accuracy for the task, particularly in the detection of very small rust disease objects on the leaves.
Article
Computer Science, Artificial Intelligence
Jorge Calvo-Zaragoza, Juan R. Rico-Juan, Antonio-Javier Gallego
Article
Computer Science, Artificial Intelligence
Javier Ortega-Bastida, Antonio Javier Gallego, Juan Ramon Rico-Juan, Pedro Albarran
Summary: This study proposes a multimodal method to predict regional GDP by combining historical GDP values with information from Twitter messages. The method, based on a two-stage architecture, successfully provides early forecasts of regional GDP and identifies the most influential opinions on the prediction.
APPLIED SOFT COMPUTING
(2021)
Article
Computer Science, Artificial Intelligence
Francisco J. Castellanos, Antonio-Javier Gallego, Jorge Calvo-Zaragoza
Summary: Binarization is a common image processing task to separate foreground and background, particularly useful for preprocessing document images. This paper proposes a method combining neural networks and Domain Adaptation to achieve unsupervised document binarization, successfully dealing with new document domains without the need for labeled data. Innovative measurement of domain similarity is used to determine the appropriateness of the adaptation process.
PATTERN RECOGNITION
(2021)
Article
Computer Science, Artificial Intelligence
Antonio Javier Gallego, Juan Ramon Rico-Juan, Jose J. Valero-Mas
Summary: The paper introduces the caKD+ algorithm which combines various techniques to improve the efficiency of kNN search, outperforming 16 state-of-the-art methods on 10 datasets.
PATTERN RECOGNITION
(2022)
Article
Computer Science, Artificial Intelligence
Antonio-Javier Gallego, Jorge Calvo-Zaragoza, Robert B. Fisher
Summary: In the context of supervised statistical learning, the study explored the issue of inconsistent distributions between training and test sets, presenting an incremental approach to address it. By utilizing an unsupervised domain adaptation algorithm to identify target samples and iteratively adapting the model through self-labeling, an adversarial training strategy was proposed to enhance the performance of domain adaptation algorithms.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2021)
Article
Automation & Control Systems
Samer Alashhab, Antonio Javier Gallego, Miguel Angel Lozano
Summary: This research proposes an interactive system for visually impaired individuals to control mobile devices using hand gestures, allowing them to perform multiple tasks without switching applications. The system utilizes a multi-head neural network and a dataset of images to detect and classify hand gestures, achieving competitive results compared to state-of-the-art methods.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Francisco J. Castellanos, Antonio Javier Gallego, Jorge Calvo-Zaragoza, Ichiro Fujinaga
Summary: Optical music recognition (OMR) is the study of automatically reading music notation from score images. Staff-region retrieval is a crucial step in the OMR workflow, but the lack of ground-truth data poses challenges. To address this, researchers propose a domain adaptation technique called Domain-Adversarial Neural Network (DANN), which shows significant improvements in F-score through experiments.
INTERNATIONAL JOURNAL ON DOCUMENT ANALYSIS AND RECOGNITION
(2022)
Article
Computer Science, Artificial Intelligence
Alejandro Sanchez-Ferrer, Jose J. Valero-Mas, Antonio Javier Gallego, Jorge Calvo-Zaragoza
Summary: The large amount of debris in the oceans has a significant impact on marine life. Efforts to tackle this problem through human-based campaigns have been insufficient due to the overwhelming amount of litter. Autonomous underwater vehicles (AUVs) have gained interest as a potential solution for locating and collecting garbage. This study explores the use of Mask Region-based Convolutional Neural Networks for automatic marine debris location and classification with limited data availability, achieving state-of-the-art results and suggesting room for further improvement.
PATTERN RECOGNITION LETTERS
(2023)
Article
Computer Science, Artificial Intelligence
Adrian Rosello, Jose J. Valero-Mas, Antonio Javier Gallego, Javier Saez-Perez, Jorge Calvo-Zaragoza
Summary: The use of deep learning in computer vision tasks can achieve remarkable results, but it depends on the availability of training data and its relationship with the application scenario. Domain adaptation techniques are crucial in robotics, where there is limited access to targeted environment data. To facilitate research in this area, Kurcuma provides a collection of datasets for kitchen utensil recognition, along with a baseline using domain-adversarial training.
PATTERN ANALYSIS AND APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Jose J. Valero-Mas, Antonio Javier Gallego, Pablo Alonso-Jimenez, Xavier Serra
Summary: This study adapts multiclass prototype generation strategies to the multilabel case and demonstrates through experiments that they significantly improve efficiency and classification performance, especially showing stronger robustness in noisy scenarios.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Information Systems
Jose J. Valero-Mas, Antonio Javier Gallego, Juan Ramon Rico-Juan
Summary: SNNs are a representative approach for Few-Shot Image Classification, utilizing weight sharing CNN models to reduce parameters and overfitting. This study assesses the representation capabilities of SNN architectures, introduces techniques such as data augmentation and transfer learning, and achieves high classification rates with limited prototypes per class.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Proceedings Paper
Automation & Control Systems
Hanz Cuevas-Velasquez, Antonio-Javier Gallego, Radim Tylecek, Jochen Hemming, Bart van Tuijl, Angelo Mencarelli, Robert B. Fisher
2020 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA)
(2020)
Article
Computer Science, Artificial Intelligence
Francisco J. Castellanos, Antonio-Javier Gallego, Jorge Calvo-Zaragoza
EXPERT SYSTEMS WITH APPLICATIONS
(2020)
Article
Computer Science, Information Systems
Antonio-Javier Gallego, Jorge Calvo-Zaragoza, Juan Ramon Rico-Juan