A comprehensive review of recent advances on deep vision systems

An improved pear disease classification approach using cycle generative adversarial network

A large number of countries worldwide depend on the agriculture, as agriculture can assist in reducing poverty, raising the country's income, and improving the food security. However, the plan diseases usually affect food crops and hence play a significant role in the annual yield and economic losses in the agricultural sector. In general, plant diseases have historically been identified by humans using their eyes, where this approach is often inexact, time-consuming, and exhausting. Recently, the employment of machine learning and deep learning approaches have significantly improved the classification and recognition accuracy for several applications. Despite the CNN models offer high accuracy for plant disease detection and classification, however, the limited available data for training the CNN model affects seriously the classification accuracy. Therefore, in this paper, we designed a Cycle Generative Adversarial Network (CycleGAN) to overcome the limitations of over-fitting and the limited size of the available datasets. In addition, we developed an efficient plant disease classification approach, where we adopt the CycleGAN architecture in order to enhance the classification accuracy. The obtained results showed an average enhancement of 7% in the classification accuracy.

HDR-EfficientNet: A Classification of Hypertensive and Diabetic Retinopathy Using Optimize EfficientNet Architecture

Hypertensive retinopathy (HR) and diabetic retinopathy (DR) are retinal diseases closely associated with high blood pressure. The severity and duration of hypertension directly impact the prevalence of HR. The early identification and assessment of HR are crucial to preventing blindness. Currently, limited computer-aided methods are available for detecting HR and DR. These existing systems rely on traditional machine learning approaches, which require complex image processing techniques and are often limited in their application. To address this challenge, this work introduces a deep learning (DL) method called HDR-EfficientNet, which aims to provide an efficient and accurate approach to identifying various eye-related disorders, including diabetes and hypertensive retinopathy. The proposed method utilizes an EfficientNet-V2 network for end-to-end training focused on disease classification. Additionally, a spatial-channel attention method is incorporated into the approach to enhance its ability to identify specific areas of damage and differentiate between different illnesses. The HDR-EfficientNet model is developed using transfer learning, which helps overcome the challenge of imbalanced sample classes and improves the network's generalization. Dense layers are added to the model structure to enhance the feature selection capacity. The performance of the implemented system is evaluated using a large dataset of over 36,000 augmented retinal fundus images. The results demonstrate promising accuracy, with an average area under the curve (AUC) of 0.98, a specificity (SP) of 96%, an accuracy (ACC) of 98%, and a sensitivity (SE) of 95%. These findings indicate the effectiveness of the suggested HDR-EfficientNet classifier in diagnosing HR and DR. In summary, the HDR-EfficientNet method presents a DL-based approach that offers improved accuracy and efficiency for the detection and classification of HR and DR, providing valuable support in diagnosing and managing these eye-related conditions.

Stacked ensembles on basis of parentage information can predict hybrid performance with an accuracy comparable to marker-based GBLUP

Testcross factorials in newly established hybrid breeding programs are often highly unbalanced, incomplete, and characterized by predominance of special combining ability (SCA) over general combining ability (GCA). This results in a low efficiency of GCA-based selection. Machine learning algorithms might improve prediction of hybrid performance in such testcross factorials, as they have been successfully applied to find complex underlying patterns in sparse data. Our objective was to compare the prediction accuracy of machine learning algorithms to that of GCA-based prediction and genomic best linear unbiased prediction (GBLUP) in six unbalanced incomplete factorials from hybrid breeding programs of rapeseed, wheat, and corn. We investigated a range of machine learning algorithms with three different types of predictor variables: (a) information on parentage of hybrids, (b) in addition hybrid performance of crosses of the parental lines with other crossing partners, and (c) genotypic marker data. In two highly incomplete and unbalanced factorials from rapeseed, in which the SCA variance contributed considerably to the genetic variance, stacked ensembles of gradient boosting machines based on parentage information outperformed GCA prediction. The stacked ensembles increased prediction accuracy from 0.39 to 0.45, and from 0.48 to 0.54 compared to GCA prediction. The prediction accuracy reached by stacked ensembles without marker data reached values comparable to those of GBLUP that requires marker data. We conclude that hybrid prediction with stacked ensembles of gradient boosting machines based on parentage information is a promising approach that is worth further investigations with other data sets in which SCA variance is high.

Mobile-HR: An Ophthalmologic-Based Classification System for Diagnosis of Hypertensive Retinopathy Using Optimized MobileNet Architecture

Hypertensive retinopathy (HR) is a serious eye disease that causes the retinal arteries to change. This change is mainly due to the fact of high blood pressure. Cotton wool patches, bleeding in the retina, and retinal artery constriction are affected lesions of HR symptoms. An ophthalmologist often makes the diagnosis of eye-related diseases by analyzing fundus images to identify the stages and symptoms of HR. The likelihood of vision loss can significantly decrease the initial detection of HR. In the past, a few computer-aided diagnostics (CADx) systems were developed to automatically detect HR eye-related diseases using machine learning (ML) and deep learning (DL) techniques. Compared to ML methods, the CADx systems use DL techniques that require the setting of hyperparameters, domain expert knowledge, a huge training dataset, and a high learning rate. Those CADx systems have shown to be good for automating the extraction of complex features, but they cause problems with class imbalance and overfitting. By ignoring the issues of a small dataset of HR, a high level of computational complexity, and the lack of lightweight feature descriptors, state-of-the-art efforts depend on performance enhancement. In this study, a pretrained transfer learning (TL)-based MobileNet architecture is developed by integrating dense blocks to optimize the network for the diagnosis of HR eye-related disease. We developed a lightweight HR-related eye disease diagnosis system, known as Mobile-HR, by integrating a pretrained model and dense blocks. To increase the size of the training and test datasets, we applied a data augmentation technique. The outcomes of the experiments show that the suggested approach was outperformed in many cases. This Mobile-HR system achieved an accuracy of 99% and an F1 score of 0.99 on different datasets. The results were verified by an expert ophthalmologist. These results indicate that the Mobile-HR CADx model produces positive outcomes and outperforms state-of-the-art HR systems in terms of accuracy.

Decoding of human identity by computer vision and neuronal vision

Extracting meaning from a dynamic and variable flow of incoming information is a major goal of both natural and artificial intelligence. Computer vision (CV) guided by deep learning (DL) has made significant strides in recognizing a specific identity despite highly variable attributes. This is the same challenge faced by the nervous system and partially addressed by the concept cells-neurons exhibiting selective firing in response to specific persons/places, described in the human medial temporal lobe (MTL) ⁠. Yet, access to neurons representing a particular concept is limited due to these neurons' sparse coding. It is conceivable, however, that the information required for such decoding is present in relatively small neuronal populations. To evaluate how well neuronal populations encode identity information in natural settings, we recorded neuronal activity from multiple brain regions of nine neurosurgical epilepsy patients implanted with depth electrodes, while the subjects watched an episode of the TV series "24". First, we devised a minimally supervised CV algorithm (with comparable performance against manually-labeled data) to detect the most prevalent characters (above 1% overall appearance) in each frame. Next, we implemented DL models that used the time-varying population neural data as inputs and decoded the visual presence of the four main characters throughout the episode. This methodology allowed us to compare "computer vision" with "neuronal vision"-footprints associated with each character present in the activity of a subset of neurons-and identify the brain regions that contributed to this decoding process. We then tested the DL models during a recognition memory task following movie viewing where subjects were asked to recognize clip segments from the presented episode. DL model activations were not only modulated by the presence of the corresponding characters but also by participants' subjective memory of whether they had seen the clip segment, and by the associative strengths of the characters in the narrative plot. The described approach can offer novel ways to probe the representation of concepts in time-evolving dynamic behavioral tasks. Further, the results suggest that the information required to robustly decode concepts is present in the population activity of only tens of neurons even in brain regions beyond MTL.

Object detection using YOLO: challenges, architectural successors, datasets and applications

Object detection is one of the predominant and challenging problems in computer vision. Over the decade, with the expeditious evolution of deep learning, researchers have extensively experimented and contributed in the performance enhancement of object detection and related tasks such as object classification, localization, and segmentation using underlying deep models. Broadly, object detectors are classified into two categories viz. two stage and single stage object detectors. Two stage detectors mainly focus on selective region proposals strategy via complex architecture; however, single stage detectors focus on all the spatial region proposals for the possible detection of objects via relatively simpler architecture in one shot. Performance of any object detector is evaluated through detection accuracy and inference time. Generally, the detection accuracy of two stage detectors outperforms single stage object detectors. However, the inference time of single stage detectors is better compared to its counterparts. Moreover, with the advent of YOLO (You Only Look Once) and its architectural successors, the detection accuracy is improving significantly and sometime it is better than two stage detectors. YOLOs are adopted in various applications majorly due to their faster inferences rather than considering detection accuracy. As an example, detection accuracies are 63.4 and 70 for YOLO and Fast-RCNN respectively, however, inference time is around 300 times faster in case of YOLO. In this paper, we present a comprehensive review of single stage object detectors specially YOLOs, regression formulation, their architecture advancements, and performance statistics. Moreover, we summarize the comparative illustration between two stage and single stage object detectors, among different versions of YOLOs, applications based on two stage detectors, and different versions of YOLOs along with the future research directions.

Classification of scanning electron microscope images of pharmaceutical excipients using deep convolutional neural networks with transfer learning

Convolutional Neural Networks (CNNs) are image analysis techniques that have been applied to image classification in various fields. In this study, we applied a CNN to classify scanning electron microscopy (SEM) images of pharmaceutical raw material powders to determine if a CNN can evaluate particle morphology. We tested 10 pharmaceutical excipients with widely different particle morphologies. SEM images for each excipient were acquired and divided into training, validation, and test sets. Classification models were constructed by applying transfer learning to pretrained CNN models such as VGG16 and ResNet50. The results of a 5-fold cross-validation showed that the classification accuracy of the CNN model was sufficiently high using either pretrained model and that the type of excipient could be classified with high accuracy. The results suggest that the CNN model can detect differences in particle morphology, such as particle size, shape, and surface condition. By applying Grad-CAM to the constructed CNN model, we succeeded in finding particularly important regions in the particle image of the excipients. CNNs have been found to have the potential to be applied to the identification and characterization of raw material powders for pharmaceutical development.

Automatic Target Detection from Satellite Imagery Using Machine Learning

Object detection is a vital step in satellite imagery-based computer vision applications such as precision agriculture, urban planning and defense applications. In satellite imagery, object detection is a very complicated task due to various reasons including low pixel resolution of objects and detection of small objects in the large scale (a single satellite image taken by Digital Globe comprises over 240 million pixels) satellite images. Object detection in satellite images has many challenges such as class variations, multiple objects pose, high variance in object size, illumination and a dense background. This study aims to compare the performance of existing deep learning algorithms for object detection in satellite imagery. We created the dataset of satellite imagery to perform object detection using convolutional neural network-based frameworks such as faster RCNN (faster region-based convolutional neural network), YOLO (you only look once), SSD (single-shot detector) and SIMRDWN (satellite imagery multiscale rapid detection with windowed networks). In addition to that, we also performed an analysis of these approaches in terms of accuracy and speed using the developed dataset of satellite imagery. The results showed that SIMRDWN has an accuracy of 97% on high-resolution images, while Faster RCNN has an accuracy of 95.31% on the standard resolution (1000 × 600). YOLOv3 has an accuracy of 94.20% on standard resolution (416 × 416) while on the other hand SSD has an accuracy of 84.61% on standard resolution (300 × 300). When it comes to speed and efficiency, YOLO is the obvious leader. In real-time surveillance, SIMRDWN fails. When YOLO takes 170 to 190 milliseconds to perform a task, SIMRDWN takes 5 to 103 milliseconds.

Deep Convolutional Neural Networks Implementation for the Analysis of Urine Culture

BACKGROUND

Urine culture images collected using bacteriology automation are currently interpreted by technologists during routine standard-of-care workflows. Machine learning may be able to improve the harmonization of and assist with these interpretations.

METHODS

A deep learning model, BacterioSight, was developed, trained, and tested on standard BD-Kiestra images of routine blood agar urine cultures from 2 different medical centers.

RESULTS

BacterioSight displayed performance on par with standard-of-care-trained technologist interpretations. BacterioSight accuracy ranged from 97% when compared to standard-of-care (single technologist) and reached 100% when compared to a consensus reached by a group of technologists (gold standard in this study). Variability in image interpretation by trained technologists was identified and annotation "fuzziness" was quantified and found to correlate with reduced confidence in BacterioSight interpretation. Intra-testing (training and testing performed within the same institution) performed well giving Area Under the Curve (AUC) ≥0.98 for negative and positive plates, whereas, cross-testing on images (trained on one institution's images and tested on images from another institution) showed decreased performance with AUC ≥0.90 for negative and positive plates.

CONCLUSIONS

Our study provides a roadmap on how BacterioSight or similar deep learning prototypes may be implemented to screen for microbial growth, flag difficult cases for multi-personnel review, or auto-verify a subset of cultures with high confidence. In addition, our results highlight image interpretation variability by trained technologist within an institution and globally across institutions. We propose a model in which deep learning can enhance patient care by identifying inherent sample annotation variability and improving personnel training.

Evolutionary Deep Attention Convolutional Neural Networks for 2D and 3D Medical Image Segmentation

Developing a convolutional neural network (CNN) for medical image segmentation is a complex task, especially when dealing with the limited number of available labelled medical images and computational resources. This task can be even more difficult if the aim is to develop a deep network and using a complicated structure like attention blocks. Because of various types of noises, artefacts and diversity in medical images, using complicated network structures like attention mechanism to improve the accuracy of segmentation is inevitable. Therefore, it is necessary to develop techniques to address the above difficulties. Neuroevolution is the combination of evolutionary computation and neural networks to establish a network automatically. However, Neuroevolution is computationally expensive, specifically to create 3D networks. In this paper, an automatic, efficient, accurate, and robust technique is introduced to develop deep attention convolutional neural networks utilising Neuroevolution for both 2D and 3D medical image segmentation. The proposed evolutionary technique can find a very good combination of six attention modules to recover spatial information from downsampling section and transfer them to the upsampling section of a U-Net-based network-six different CT and MRI datasets are employed to evaluate the proposed model for both 2D and 3D image segmentation. The obtained results are compared to state-of-the-art manual and automatic models, while our proposed model outperformed all of them.

A Fuzzy Shell for Developing an Interpretable BCI Based on the Spatiotemporal Dynamics of the Evoked Oscillations

Researchers in neuroscience computing experience difficulties when they try to carry out neuroanalysis in practice or when they need to design an explainable brain-computer interface (BCI) with quick setup and minimal training phase. There is a need of interpretable computational intelligence techniques and new brain states decoding for more understandable interpretation of the sensory, cognitive, and motor brain processing. We propose a general-purpose fuzzy software system shell for developing a custom EEG BCI system. It relies on the bursts of the ongoing EEG frequency power synchronization/desynchronization at scalp level and supports quick BCI setup by linguistic features, ad hoc fuzzy membership construction, explainable IF-THEN rules, and the concept of the Internet of Things (IoT), which makes the BCI system device and service independent. It has a potential for designing both passive and event-related BCIs with options for visual representation at scalp-source level in response to time. The feasibility of the proposed system has been proven by real experiments and bursts for β and γ frequency power have been detected in real time in response to evoked visuospatial selective attention. The presence of the proposed new brain state decoding can be used as a feasible metric for interpretation of the spatiotemporal dynamics of the passive or evoked neural oscillations.

Automatic identification of cashmere and wool fibers based on microscopic visual features and residual network model

Distinguishing cashmere and sheep wool fibers is a challenge. In this study, we propose a residual net-based method for the identification of cashmere and sheep wool fibers. First, optical microscopic images of six different types of cashmere and sheep wool fibers were collected, and then the sample images were data-augmented. Several classic convolutional neural network (CNN) models were trained and tested with the sample images. The comparison showed that the proposed residual net model with 18 weight layers had the highest accuracy, with an overall accuracy above 97.1 % on the test set; the highest accuracy on the Australian merino wool and Mongolian brown cashmere, both above 98 %; and the lowest accuracy on the Chinese white cashmere, above 95 %. The trained model exhibited a fast detection speed, processing 6000 sample images in less than 20 s.

Driver Fatigue Detection Systems Using Multi-Sensors, Smartphone, and Cloud-Based Computing Platforms: A Comparative Analysis

Internet of things (IoT) cloud-based applications deliver advanced solutions for smart cities to decrease traffic accidents caused by driver fatigue while driving on the road. Environmental conditions or driver behavior can ultimately lead to serious roadside accidents. In recent years, the authors have developed many low-cost, computerized, driver fatigue detection systems (DFDs) to help drivers, by using multi-sensors, and mobile and cloud-based computing architecture. To promote safe driving, these are the most current emerging platforms that were introduced in the past. In this paper, we reviewed state-of-the-art approaches for predicting unsafe driving styles using three common IoT-based architectures. The novelty of this article is to show major differences among multi-sensors, smartphone-based, and cloud-based architectures in multimodal feature processing. We discussed all of the problems that machine learning techniques faced in recent years, particularly the deep learning (DL) model, to predict driver hypovigilance, especially in terms of these three IoT-based architectures. Moreover, we performed state-of-the-art comparisons by using driving simulators to incorporate multimodal features of the driver. We also mention online data sources in this article to test and train network architecture in the field of DFDs on public available multimodal datasets. These comparisons assist other authors to continue future research in this domain. To evaluate the performance, we mention the major problems in these three architectures to help researchers use the best IoT-based architecture for detecting DFDs in a real-time environment. Moreover, the important factors of Multi-Access Edge Computing (MEC) and 5th generation (5G) networks are analyzed in the context of deep learning architecture to improve the response time of DFD systems. Lastly, it is concluded that there is a research gap when it comes to implementing the DFD systems on MEC and 5G technologies by using multimodal features and DL architecture.

Current Status, Challenges, and Possible Solutions of EEG-Based Brain-Computer Interface: A Comprehensive Review

Brain-Computer Interface (BCI), in essence, aims at controlling different assistive devices through the utilization of brain waves. It is worth noting that the application of BCI is not limited to medical applications, and hence, the research in this field has gained due attention. Moreover, the significant number of related publications over the past two decades further indicates the consistent improvements and breakthroughs that have been made in this particular field. Nonetheless, it is also worth mentioning that with these improvements, new challenges are constantly discovered. This article provides a comprehensive review of the state-of-the-art of a complete BCI system. First, a brief overview of electroencephalogram (EEG)-based BCI systems is given. Secondly, a considerable number of popular BCI applications are reviewed in terms of electrophysiological control signals, feature extraction, classification algorithms, and performance evaluation metrics. Finally, the challenges to the recent BCI systems are discussed, and possible solutions to mitigate the issues are recommended.