Search

We propose an alternative to the generative classifier that usually models both the class conditionals and class priors separately, and then uses the Bayes theorem to compute the posterior distribution of classes given the training set as a decision boundary. Because SVM (support vector machine) is not a probabilistic framework, it is really difficult to implement a direct posterior distribution-based discriminative classifier. As SVM lacks in full Bayesian analysis, we propose a hybrid (generative–discriminative) technique where the generative topic features from a Bayesian learning are fed to the SVM. The standard latent Dirichlet allocation topic model with its Dirichlet (Dir) prior could be defined as Dir–Dir topic model to characterize the Dirichlet placed on the document and corpus parameters. With very flexible conjugate priors to the multinomials such as generalized Dirichlet (GD) and Beta-Liouville (BL) in our proposed approach, we define two new topic models: the BL–GD and GD–BL. We take advantage of the geometric interpretation of our generative topic (latent) models that associate a K-dimensional manifold (K is the size of the topics) embedded into a V-dimensional feature space (word simplex) where V is the vocabulary size. Under this structure, the low-dimensional topic simplex (the subspace) defines a document as a single point on its manifold and associates each document with a single probability. The SVM, with its kernel trick, performs on these document probabilities in classification where it utilizes the maximum margin learning approach as a decision boundary. The key note is that points or documents that are close to each other on the manifold must belong to the same class. Experimental results with text documents and images show the merits of the proposed framework.

Phenology has become a field of growing importance due to the increasingly apparent impacts of climate change. However, the time-consuming, subjective and tedious nature of traditional human field observations have hindered the development of large-scale phenology networks. Such networks are rare and rely on time-lapse cameras and simplistic color indexes to monitor phenology. To automatize rapid, detailed and repeatable analyzes, we propose an Artificial Intelligence (AI) framework based on machine learning and computer vision techniques. Our approach extracts multiple ecologically-relevant indicators from time-lapse digital photography datasets. The proposed framework consists of three main components: (i) a random forest model to automatically select relevant images based on color information; (ii) a convolutional neural network (CNN) to identify and localize open tree buds; and (iii) a density-based spatial clustering algorithm to cluster open bud detections across the time-series. We tested this framework on a dataset including thousands of black spruce and balsam fir tree images captured using our phenological camera network. The performed experiments showed the efficiency of the proposed approach under challenging perturbation factors, such as significant image noise. Our framework is exceedingly faster and more accurate than human analysts, reducing the time-series processing time from multiple days to under an hour. The proposed methodology is particularly appropriate for large-scale and long-term analyzes of ecological imagery datasets. Our work demonstrates that the use of computer vision and machine learning methods represents a promising direction for the implementation of national, continental, or even global plant phenology networks.

Mechanical site preparation by mounding is often used by the forest industry to provide optimal growth conditions for tree seedlings. Prior to planting, an essential step consists in estimating the number of mounds at each planting block, which serves as planting microsites. This task often requires long and costly field surveys, implying several forestry workers to perform manual counting procedure. This paper addresses the problem of automating the counting process using computer vision and UAV imagery. We present a supervised detection-based counting framework for estimating the number of planting microsites on a mechanically prepared block. The system is trained offline to learn feature representations from semi-automatically annotated images. Mound detection and counting are then performed on multispectral UAV images captured at an altitude of 100 m. Our detection framework proceeds by generating region proposals based on local binary patterns (LBP) features extracted from near-infrared (NIR) patches. A convolutional neural network (CNN) is then used for classifying candidate regions by considering multispectral image data. To train and evaluate the proposed method, we constructed a new dataset by capturing aerial images from different planting blocks. The results demonstrate the efficiency and validity of the proposed method under challenging experimental conditions. The methods and results presented in this paper form a promising cornerstone to develop advanced decision support systems for planning planting operations.

Assessment of respiratory activity in pediatric intensive care unit allows a comprehensive view of the patient’s condition. This allows the identification of high-risk cases for prompt and appropriate medical treatment. Numerous research works on respiration monitoring have been conducted in recent years. However, most of them are unsuitable for clinical environment or require physical contact with the patient, which limits their efficiency. In this paper, we present a novel system for measuring the breathing pattern based on a computer vision method and contactless design. Our 3D imaging system is specifically designed for pediatric intensive care environment, which distinguishes it from the other imaging methods. Indeed, previous works are mostly limited to the use of conventional video acquisition devices, in addition to not considering the constraints imposed by intensive care environment. The proposed system uses depth information captured by two (Red Green Blue-Depth) RGB-D cameras at different view angles, by considering the intensive care unit constraints. Depth information is then exploited to reconstruct a 3D surface of a patient's torso with high temporal and spatial resolution and large spatial coverage. Our system captures the motion information for the top of the torso surface as well as for its both lateral sides. For each reconstruction, the volume is estimated through a recursive subdivision of the 3D space into cubic unit elements. The volume change is then calculated through a subtraction technique between successive reconstructions. We tested our system in the pediatric intensive care unit of the Sainte-Justine university hospital center, where it was compared to the gold standard method currently used in pediatric intensive care units. The performed experiments showed a very high accuracy and precision of the proposed imaging system in estimating respiratory rate and tidal volume.

Siamese trackers demonstrated high performance in object tracking due to their balance between accuracy and speed. Unlike classification-based CNNs, deep similarity networks are specifically designed to address the image similarity problem, and thus are inherently more appropriate for the tracking task. However, Siamese trackers mainly use the last convolutional layers for similarity analysis and target search, which restricts their performance. In this paper, we argue that using a single convolutional layer as feature representation is not an optimal choice in a deep similarity framework. We present a Multiple Features-Siamese Tracker (MFST), a novel tracking algorithm exploiting several hierarchical feature maps for robust tracking. Since convolutional layers provide several abstraction levels in characterizing an object, fusing hierarchical features allows to obtain a richer and more efficient representation of the target. Moreover, we handle the target appearance variations by calibrating the deep features extracted from two different CNN models. Based on this advanced feature representation, our method achieves high tracking accuracy, while outperforming the standard siamese tracker on object tracking benchmarks. The source code and trained models are available at https://github.com/zhenxili96/MFST., Accepted for Machine Vision and Applications, 2021. arXiv admin note: substantial text overlap with arXiv:2103.00810

Siamese trackers have recently achieved interesting results due to their balance between accuracy and speed. This success is mainly due to the fact that deep similarity networks were specifically designed to address the image similarity problem. Therefore, they are inherently more appropriate than classical CNNs for the tracking task. However, Siamese trackers rely on the last convolutional layers for similarity analysis and target search, which restricts their performance. In this paper, we argue that using a single convolutional layer as feature representation is not the optimal choice within the deep similarity framework, as multiple convolutional layers provide several abstraction levels in characterizing an object. Starting from this motivation, we present the Multi-Features Siamese Tracker (MFST), a novel tracking algorithm exploiting several hierarchical feature maps for robust deep similarity tracking. MFST proceeds by fusing hierarchical features to ensure a richer and more efficient representation. Moreover, we handle appearance variation by calibrating deep features extracted from two different CNN models. Based on this advanced feature representation, our algorithm achieves high tracking accuracy, while outperforming several state-of-the-art trackers, including standard Siamese trackers. The code and trained models are available at https://github.com/zhenxili96/MFST., Comment: ICPR 2021, Oral