Accurate and timely diagnosis of plant diseases plays a crucial role in ensuring global food security, safeguarding human health, and supporting the economic well-being of farmers. Currently, there is a growing interest in developing reliable deep learning systems for identifying various plant species and their associated diseases. In this respect, combining multiple CNN is an effective approach for improving disease classification since each individual model has unique design principles that enable them to extract different levels and types of features from input images. The present work introduces a novel fuzzy aggregation-based CNN ensemble for tomato leaf disease classification. Initially, five CNN architectures, namely, VGG16, ResNet152V2, MobileNetV2, InceptionV3, and DenseNet201, are designed for classifying tomato leaf diseases. Based on their diverse statistical characteristics, selected models are then integrated into a CNN ensemble. To achieve this step, we introduce a new fuzzy Max-Sum integral, built upon a fuzzy convex combination operator. Performance assessment is conducted on two datasets including 10 and 6 tomato disease classes. Further experiments are conducted by merging these datasets to include a broader range of disease categories. The results demonstrate that the proposed CNN ensemble outperforms both individual models and the CNN ensemble based on the classical Sugeno fuzzy integral.

Weed recognition is an essential step for automatic weed control systems. Identifying weeds enables targeted control measures to be implemented, minimizing the use of chemicals and reducing the impact on the environment. Deep learning-based approaches proved to be effective for addressing various complex classification problems. However, to benefit fully from their capabilities, large amounts of labeled data are required, which represents a limitation for agricultural applications, consequence of the tedious and time-consuming process of data labeling. Conversely, unlabeled data could be acquired in large quantities, with relative ease. Hence, our aim is to develop robust and precise deep learning models, to carry-out the recognition and identification of weed species, using both types of data. To this end, we propose a method, that adopts the semi-supervised learning paradigm, to optimally combine labeled and unlabeled data. The method is based on a new deep neural networks architecture, which consists of a modernized convolutional encoder belonging to the family ConvNeXt and a thoroughly designed deep decoder network. This architecture, enables a successful integration of consistency regularization. The conducted experiments on DeepWeeds and 4-Weeds, showed that the semi-supervised models trained through our proposed method provide a stable and high classification performance, compared to other state-of-the-art deep learning models, which were affected negatively by the amount of labeled data available, and the presence of noise during inference. Furthermore, the effectiveness of the proposed method was demonstrated in comparison to other semi-supervised learning methods. The results obtained demonstrate the benefits of adopting the semi-supervised learning paradigm, especially in scenarios with very limited labeled data.

Tensor decomposition has been demonstrated to be successful in a wide range of applications, from neuroscience and wireless communications to social networks. In an online setting, factorizing tensors derived from multidimensional data streams is however nontrivial due to several inherent problems of real-time stream processing. In recent years, many research efforts have been dedicated to developing online techniques for decomposing such tensors, resulting in significant advances in streaming tensor decomposition or tensor tracking. This topic is emerging and enriches the literature on tensor decomposition, particularly from the data stream analystics perspective. Thus, it is imperative to carry out an overview of tensor tracking to help researchers and practitioners understand its developments and achievements, summarise the current trends and advances, and identify challenging problems. In this article, we provide a contemporary and comprehensive survey on different types of tensor tracking techniques. We particularly categorize the state-of-the-art methods into three main groups: streaming CP decompositions, streaming Tucker decompositions, and streaming decompositions under other tensor formats (i.e., tensor-train, t-SVD, and BTD). In each group, we further divide the existing algorithms into sub-categories based on their main optimization framework and model architectures. Finally, we present several applications, research challenges, open problems, and potential directions of tensor tracking in the future.

We studied the problem of sparse subspace tracking in the high-dimensional regime where the dimension is comparable to or much larger than the sample size. Leveraging power iteration and thresholding methods, a new provable algorithm called OPIT was derived for tracking the sparse principal subspace of data streams over time. We also presented a theoretical result on its convergence to verify its consistency in high dimensions. Several experiments were carried out on both synthetic and real data to demonstrate the effectiveness of OPIT.

Most vision-language tasks, such as image captioning, require complex deep learning (DL) models/architectures. However, the high complexity of such models often leads their inner functioning to be regarded as black-boxes, preventing humans from understanding their decision process. Representation space is a key element in DL models, understanding this space and how information is encoded constitute an interesting approach for interpretable DL. Indeed, as far as we are aware, this paper is the first one presenting a novel method based on perturbation principle of representation space, to study the components that influence DL architectures used in image captioning. The core idea is to isolate and identify the importance of each component (element of the architecture) involved in the captioning pipeline by perturbing, by means of Gaussian functions, the representation space rather than the original space of inputs. We experimentally demonstrate that those components differ in their influence and relevance. The results show that the visual modality would constitute a critical explanation target in captioning models in contrast to language modality, thus leading to more fine-grained explanations. We also propose MSICE, an automatic evaluation metric for image captioning which addresses two important yet overlooked linguistic aspects, morphology and semantic. Our code will be publicly accessible to support future research in this field.

In this paper, we introduce a fast adaptive algorithm for CAN- DECOMP/PARAFAC decomposition of streaming three-way tensors using randomized sketching techniques. By leveraging randomized least-squares regression and approximating matrix multiplication, we propose an efficient first-order estimator to minimize an exponentially weighted recursive least- squares cost function. Our algorithm is fast, requiring a low computational complexity and memory storage. Experiments indicate that the proposed algorithm is capable of adaptive tensor decomposition with a competitive performance evaluation on both synthetic and real data.

The present work focuses on combining Convolutional Neural Network (CNN) features to strengthen the writer retrieval in historical document databases. Various CNN models that are LeNet, ResNet, and VGG are used to get writer-independent features of handwritten documents. These features are associated with dissimilarity measures to achieve the retrieval task. Then, developed writer retrieval systems are combined through SVM classifier. Experiments are conducted on ICDAR-2017 dataset which contains historical handwritten documents. The results obtained highlight the robustness of CNN features and the combination stage as well.

Ultrasound-Guided Regional Anesthesia (UGRA) becomes a standard procedure in surgical operations and contributes to pain management. It offers the advantages of the targeted nerve detection and provides the visualization of regions of interest such as anatomical structures. However, nerve detection is one of the most challenging tasks that anesthetists can encounter in the UGRA procedure. A computer-aided system that can detect automatically the nerve region would facilitate the anesthetist's daily routine and allow them to concentrate more on the anesthetic delivery. In this paper, we propose a new method based on merging deep learning models from different data to detect the median nerve. The merged architecture consists of two branches, one being one-dimensional (1D) convolutional neural networks (CNN) branch and another 2D CNN branch. The merged architecture aims to learn the high-level features from 1D handcrafted noise-robust features and 2D ultrasound images. The obtained results show the validity and high accuracy of the proposed approach and its robustness.

Texture is an important characteristic for different computer vision tasks and applications. Local binary pattern (LBP) is considered one of the most efficient texture descriptors yet. However, LBP has some notable limitations, in particular its sensitivity to noise. In this paper, we address these criteria by introducing a novel texture descriptor, robust adaptive median binary pattern (RAMBP). RAMBP is based on a process involving classification of noisy pixels, adaptive analysis window, scale analysis, and a comparison of image medians. The proposed method handles images with highly noisy textures and increases the discriminative properties by capturing microstructure and macrostructure texture information. The method was evaluated on popular texture datasets for classification and retrieval tasks and under different high noise conditions. Without any training or prior knowledge of the noise type, RAMBP achieved the best classification compared to state-of-the-art techniques. It scored more than 90% under 50% impulse noise densities, more than 95% under Gaussian noised textures with a standard deviation σ = 5 , more than 99% under Gaussian blurred textures with a standard deviation σ = 1.25 , and more than 90% for mixed noise. The proposed method yielded competitive results and proved to be one of the best descriptors in noise-free texture classification. Furthermore, RAMBP showed high performance for the problem of noisy texture retrieval providing high scores of recall and precision measures for textures with high noise levels. Finally, compared with the state-of-the-art methods, RAMBP achieves a good running time with low feature dimensionality.

Automatic image classification systems for indirect immunofluorescence (IIF) labeling of human epithelial (HEp-2) cell specimens are needed to improve the efficient management of autoimmune diseases. In this paper, we propose to classify HEp-2 cell specimen imagery using a combination of local features and deep learning features extracted from the IIF images. Two local descriptors are used to capture texture information, namely: Rotation Invariant Co-occurrence among Local Binary Patterns (RIC-LBP) extending the LBP descriptor and Joint Motif Labels (JML) based on the Peano scan motif concept. Deep learning features are then extracted using the VGG-19 image classification network. Finally, all descriptors are combined using a late fusion approach with a Random Forests (RF) classifier with seven output classes. Experimental results show that our proposed framework achieves a mean class accuracy of 92.11% with five-fold cross validation using the RF classifier with 1000 trees on the HEp-2 specimen benchmark dataset, which outperforms the state-of-the-art accuracy on this dataset.

The leading advantage of Ultrasound-Guided Regional Anesthesia (UGRA) is the ability to visualize the anatomical structure of interest. The nerve detection should translate into greater efficacy, by ensuring accurate deposition and spread of local anesthetic around the target nerve. The identification of nerve is among the most difficult tasks that anesthetists can encounter in the UGRA procedure. It is important to develop an automatic nerve detection algorithm to assist the UGRA practitioners. However, robust and automated nerve detection is challenging due to inherent noise and artifacts in ultrasound images and to the variability of the nerve structure. In order to deal with such a problem we propose a new method based on learning and estimating a dynamic on-line model of nerve position consistency and confidence measure by using generalized Gaussian distribution assumption. Based on that, Naive Bayesian measure is used to indicate whether a detected nerve region is reliable or not. Experiments were conducted on 7000 ultrasound images from 10 patients achieving an average f-score of 83%.

Human epithelial (HEp-2) cell specimens are obtained from indirect immunofluorescence (IIF) imaging for diagnosis and management of autoimmune diseases. Analysis of HEp2 cells is important and in this work we consider automatic cell segmentation and classification using spatial and texture pattern features and random forest classifiers. In this paper, we summarize our efforts in classification and segmentation tasks proposed in ICPR 2016 contest. For the cell level staining pattern classification (Task 1), we utilized texture features such as rotational invariant co-occurrence (RIC) versions of the well-known local binary pattern (LBP), median binary pattern (MBP), joint adaptive median binary pattern (JAMBP), and motif labels (ML) along with other optimized features. We report the classification results utilizing different classifiers such as the k-nearest neighbors (kNN), support vector machine (SVM), and random forest (RF). We obtained the best accuracy of 94.26% for six cell classes with RIC-LBP combined with a motif pattern co-occurrence labels (MCL). For specimen level staining pattern classification (Task 2) we utilize a combination RIC-LBP with RF classifier and obtain 80% accuracy for seven classes. For cell segmentation (Task 4), we use our optimized multiscale spatial feature bank along with RF classifier for pixel-wise labeling to achieve an F-measure of 84.26% for 1008 images.

It is increasingly appreciated that the tumor stroma is an integral part of cancer initiation, growth, and progression. Recently it has been shown that the stromal elements of tumors hold prognostic as well as response-predictive information. This work proposes a multi-scale image analysis and machine learning pipeline for epithelial versus stromal tissue identification in images of H&E stained breast cancer specimens. Unlike many studies that perform pixel or block-based epithelium-stroma classification, this pipeline includes an explicit image segmentation module. We first partition the H&E stain images into coherent partitions/superpixels, then extract a number of regional color and texture features from these partitions, and finally use support vector machine classifiers to classify them into epithelium and stroma classes. We propose a multi-scale hierarchical fuzzy c-means (HFCM) approach for segmentation of the images. We also investigate multi-scale feature extraction and descriptors. Our experimental results on Stanford Tissue Microarray Database show that multi-scale regional feature descriptors outperform single-scale feature descriptors. Experimental results also show that when the same set of regional features are used, classification of HFCM-based partitions outperforms classification of both regular blocks and SLIC superpixels.

Ultrasound-guided regional anaesthesia (UGRA) is growing rapidly in the medical field, and becomes a standard procedure in many worldwide hospitals. UGRA can specifically benefit from image processing and machine learning techniques. Very few studies have been developed for that purpose. This paper focuses on automatic localization of nerve in ultrasound images, in order to assist anaesthetists during UGRA procedure. Due to the complex structure of nerve and poor quality of ultrasound images, the automatic detection of nerve region is a challenging problem. To handle such issue, several processing phases are required. For that purpose, we propose a new method, based on despeckling, feature ranking and majority vote classification, for a robust and accurate median nerve localization. The proposed method is applied on a real dataset obtained from eight patients. The obtained results showed high performance for median nerve detection achieving accuracy of 89% of the f-score measure.

This paper addresses the challenging problem of recognition and classification of textured surfaces under illumination variation, geometric transformations and noisy sensor measurements. We propose a new texture operator, Adaptive Median Binary Patterns (AMBP) that extends our previous Median Binary Patterns (MBP) texture feature. The principal idea of AMBP is to hash small local image patches into a binary pattern text on by fusing MBP and Local Binary Patterns (LBP) operators combined with using self-adaptive analysis window sizes to better capture invariant microstructure information while providing robustness to noise. The AMBP scheme is shown to be an effective mechanism for non-parametric learning of spatially varying image texture statistics. The local distribution of rotation invariant and uniform binary pattern subsets extended with more global joint information are used as the descriptors for robust texture classification. The AMBP is shown to outperform recent binary pattern and filtering-based texture analysis methods on two large texture corpora (CUReT and KTH_TIPS2-b) with and without additive noise. The AMBP method is slightly superior to the best techniques in the noiseless case but significantly outperforms other methods in the presence of impulse noise.

Forest fires constitute one of the most damaging natural disaster for many countries around the world. Its mechanisms can be studied through forest fire metrology. Despite a large number of proposed algorithms for fire detection, only few works adress the segmentation problem in fire metrology. The main purpose of this paper is to introduce a SVM-based segmentation method for forest fire metrology. This method is confronted with state-of-the-art fire segmentation algorithms using supervised evaluation criteria and an ad hoc expertised picture database. The obtained results highlight the good performances of the proposed method compared to prior algorithms.

Computer assisted or automated histological grading of tissue biopsies for clinical cancer care is a long-studied but challenging problem. It requires sophisticated algorithms for image segmentation, tissue architecture characterization, global texture feature extraction, and high-dimensional clustering and classification algorithms. Currently there are no automatic image-based grading systems for quantitative pathology of cancer tissues. We describe a novel approach for tissue segmentation using fuzzy spatial clustering, vector-based multiphase level set active contours and nuclei detection using an iterative kernel voting scheme that is robust even in the case of clumped touching nuclei. Early results show that we can reach a 91% detection rate compared to manual ground truth of cell nuclei centers across a range of prostate cancer grades.

We present a new approach aimed at fast multiframe registration of airborne video collected by moving platforms such as unmanned aerial vehicles. Registration is used to enable separating the moving objects from the stationary background, which is similar to estimating the egomotion of the sensor. The proposed registration algorithm is used to match the moving background and remap video frames into a common co-ordinate system in order to stabilize that segment of video. The major modules include dense feature detection, sparse prominent edge and corner-based feature block identification, feature block region matching to extract control points, confidence weighted robust projective transformation estimation and image warping to register a segment of frames within a given temporal window. The proposed method is shown to produce good results with small image registration error.

Local information provides a useful and efficient means for analyzing texture. The classification performance depends on how this information is characterized. In this paper we propose a novel method based on relative relations to locally describe regions in the image. This method presents many advantages: low complexity, illumination invariance and powerful discriminative properties. Using a set of classifications we show that the proposed technique improves significantly the capability of texture recognition and out performs classical methods.

This paper describes a new method for fuzzy segmentation based on spatial constraints. Taking into account the neighborhood influence two techniques are used. First a new feature is derived from Peano scans to represent a spatial relationship among neighbors. Second we incorporate a regularization term to fuzzy C-means algorithm (FCM). The algorithm is tested on both synthetic and multispectral images. Experimental results are presented and discussed. They show the effectiveness of the method.