Deep Learning Processing Research Articles

A Deep Learning Approach Based on Physical Constraints for Predicting Soil Moisture in Unsaturated Zones

AbstractWater transport in the unsaturated zone is an important part of the hydrological cycle and is the link between the atmosphere‒soil‐groundwater for material and energy transport. The accurate prediction of soil moisture (SM) is essential for the rational exploitation of water resources. Data‐driven deep learning methods are widely used in many fields; however, the lack of physical mechanisms limits their application in hydrological fields, especially for SM prediction in unsaturated zones. To solve this problem, this study proposes a new deep learning method that introduces the water balance principle, Richard's equation, and SM boundary conditions as constraints to construct the new loss function that guides the training process of deep learning, called physics‐informed deep learning (PIDL). In tests consisting of a large number of data sets acquired from in situ observation sites in the field, PIDL exhibits higher accuracy than ordinary deep learning (long short‐term memory) and physical models, with 51.03% and 53.46% reduction in root mean square error of SM prediction, respectively. PIDL performance significantly improved in predicting scenarios that are difficult for ordinary deep learning to handle, such as sparse data sets, extreme values, and mutated values. In addition, PIDL maintains high accuracy over a longer prediction period. The addition of physical mechanisms allows deep learning to mine patterns not only from the data itself but also from a priori physical theoretical knowledge for guidance, and this hybrid modeling approach can also be generalized to prediction problems in other hydrological domains.

RESDEN: A Novel Deep Unified Model for Face Recognition System

The Face Recognition technology plays a significant role in the field of Computer Vision in contemporary times. The research article is centered on a Facial attendance system that utilizes a deep learning technique to recognize face photos. To execute face identification and classification via the use of deep learning processes, many Convolutional Neural Network (CNN) models are taken into account. Previous studies have mostly focused on either the ResNet or DenseNet-based convolutional neural network model. The present research utilizes the merging of ResNet and DenseNet to propose a hybrid model. The proposed work is expected to provide enhanced efficiency and accuracy. In the training and testing stages of the simulation, considerations are made for both binary and category classifications. The current research focuses on the use of the LFW dataset. The pictures undergo an initial step of the noise reduction process. The evaluation of picture quality is conducted by taking into account metrics such as Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index (SSIM). After the proposed model has undergone training, it generates photographs of superior quality. Finally, the proposed system incorporates the RESDEN framework, which integrates DenseNet with a noise reduction technique, a segmentation mechanism, and a CNN based on ResNet. A comparative analysis has been conducted to evaluate the accuracy of several filtered picture sets across different convolutional neural network (CNN) models. The simulation results indicate that the suggested model exhibited a good level of performance and accuracy.

A‐PGRD: Attention‐based automatic business process model generation from RPA process description

SummaryRobotic process automation (RPA), a tool driven by business processes as the kernel, continues to heat up in the business community. However, process‐centric RPA modeling lacks an effective means. To address this problem, we propose a method for automatic process acquisition using RPA process descriptions as input. Existing deep learning process generation methods cannot be applied at the phrase level and have low accuracy at the sentence level. The proposed neural network method is based on an attention mechanism for automatic business process model generation from RPA process descriptions (A‐PGRD). The approach analyzes easily accessible and unstructured natural language text documents, constructs a non‐autoregressive neural network with an attention mechanism to retrieve the business process hierarchy, and generates a tree‐like business process graph using unsupervised automation. Through K‐fold cross‐validation, the method achieves an accuracy of 41.7% on the manually collected open‐source RPA business process dataset. Compared with the previous method, the method improves the learning efficiency by 23%–27%. The obtained results can be applied to the RPA tool to better optimize the business process and thus help organizations gain an edge over their competition.

Analyze IMDb movies by sentiment and topic analysis

Movie is an important cultural form, carrying multiple levels and meanings such as art, entertainment and social value. Movie review and rating data sets are huge, and deep learning and natural language processing methods are widely used today. Advances in big data and deep learning offer unprecedented opportunities to understand moviegoer behavior and preferences while providing a cost-effective way to gain insights relevant to the entertainment industry. This project conducts sentiment analysis, topic modeling, and visual statistical analysis based on the IMDb movie data set to identify key factors and deeper insights that influence successful decision-making in film production. This project first uses the word embedding method to vectorize the movie review text, and then uses Bidirectional Long Short-Term Memory (Bi-LSTM) to perform sentiment classification. In addition, statistical methods such as visualization were used to discover conclusions such as the highest average number of movies released in November, and identify trends, patterns and relationships between the variables of IMDb movies. Finally, the Latent Dirichlet Allocation (LDA) topic modeling model was constructed to find out that the important topic with increased demand is light entertainment movies, highlighting the commercial feasibility of comedy movies as a profitable business model. In summary, this project uses an emotion-topic fusion analysis method based on the Bi-LSTM emotion classification method and the LDA topic modeling method. The results show that the Bi-LSTM model can better identify positive and negative emotions in movie reviews, and the LDA topic model performs well in mining popular topics.

Review on ultrasonographic diagnosis of thyroid diseases based on deep learning

In recent years, the incidence of thyroid diseases has increased significantly and ultrasound examination is the first choice for the diagnosis of thyroid diseases. At the same time, the level of medical image analysis based on deep learning has been rapidly improved. Ultrasonic image analysis has made a series of milestone breakthroughs, and deep learning algorithms have shown strong performance in the field of medical image segmentation and classification. This article first elaborates on the application of deep learning algorithms in thyroid ultrasound image segmentation, feature extraction, and classification differentiation. Secondly, it summarizes the algorithms for deep learning processing multimodal ultrasound images. Finally, it points out the problems in thyroid ultrasound image diagnosis at the current stage and looks forward to future development directions. This study can promote the application of deep learning in clinical ultrasound image diagnosis of thyroid, and provide reference for doctors to diagnose thyroid disease.

Disintermediation and disinformation as a political strategy: use of AI to analyse fake news as Trump’s rhetorical resource on Twitter

The communicative effects of disintermediation caused by social media promote the expansion of personalist and emotional political discourses that reach the audience directly and evade the traditional journalistic filter. This phenomenon leads to new political communication tactics, but also exposes citizens to potentially fraudulent, contaminated or polarised content. In this context, framed in post-truth, the term ‘fake news’ gains relevance as a way of referring to disinformation and as a political and performative argument that can be weaponised. This research aims to analyse such use in the discourse of the former president Donald Trump during his presidential term (2017-2021), focussing on Twitter as the main platform in his political communication strategy online. To analyse this, we resort to a methodological triangulation of content, discourse, and sentiment analysis, with the latter combining both lexicon and artificial intelligence (AI) techniques through machine learning on the basis of deep learning and natural language processing, which is applied to his messages published with the term ‘fake news’ (N = 768). The analysis of the sample, provided here in an open dataset, employs self-developed software that allows each unit of analysis to be filtered and coded around its predominant themes, sentiments, and words. The main results confirm that Trump’s attribution of ‘fake news’ focusses on three main topics: the media (53%), politics (40%) and his cabinet (33%). It also shows how the former president resorts to a personalist agenda, focussed on the defence of his proposals and his team (80%) by delegitimizing his opponents and the press, with a negative tone (72%) loaded with derogatory terms, confirming a weaponised strategy of the term ‘fake news’ as a political argument of disinformation and disintermediation.

Prediction of intensified ethanol fermentation of sugarcane using a deep learning soft sensor and process analytical technology

AbstractBACKGROUNDIntensified ethanol fermentation produces higher ethanol concentrations while reducing water and energy requirements. Nevertheless, the inhibitory and detrimental effect of the cellular stress barriers in this process further complicates the nonlinear dynamic relationship between the variables that directly reflect the fermentation quality. These key variables are hard to measure in real time and therefore cannot be directly controlled.RESULTSThis work presents the development of a soft sensor that predicts in real time the ethanol and substrate concentrations of an intensified fermentation. The soft sensor uses feedforward neural networks (FNNs) with easily measurable process analytical technology (PAT) tools. The application of advanced PAT tools such as redox potential and capacitance, in addition to temperature and pH are explored as input variables. The complex kinetic relationship between the studied variables was captured with FNN architectures with a single hidden layer and between 95 and 175 hidden neurons for the different cases studied. Acceptable predictions are achieved for the concentration of ethanol (RMSE = 9.5 and R2 = 0.97) and substrate (RMSE = 17.02 and R2 = 0.92).CONCLUSIONSThe results confirm that the proposed soft sensor can accurately predict the ethanol and substrate concentrations. Collectively, capacitance, redox potential, temperature and pH provide a powerful platform of PAT tools that can directly infer key variables showing the fermentation quality in real time. This study provides a significant step towards the systematic development of a reliable soft sensor with integration of advanced PAT tools. © 2023 Society of Chemical Industry.

A review of plant disease detection and classification methods

ABSTRACT The majority of agricultural yields are affected by plant diseases leading to economic losses. Hence, it is necessary to evaluate plant diseases at the very beginning to enhance crop production by making effective control actions. Therefore, automatic plant disease detection models are required to reduce the huge laborious tasks. In recent times, deep learning and computer vision models have been established for the automated detection of plant disease which functions by analyzing the symptoms on plant leaves. Thus, this paper reviews a set of articles that utilizes deep learning and image processing in the field of detecting plant and crop diseases. This survey ensures a comprehensive view of all research works regarding plant disease detection for exploring the effectiveness of the existing models for detecting diseases in diverse crops. This paper aims to analyze various machine learning and deep learning approaches regarding crop disease identification. This review gives an introduction to plant leaf diseases, the conventional methods for detecting plant diseases, the dataset used, the performance metrics utilized for analyzing the existing detection frameworks, and the implementation platforms. Finally, this review provides the limitations in this field of plant leaf detection and the future research trends for further enhancement.

Deep reinforcement transfer learning of active control for bluff body flows at high Reynolds number

We demonstrate how to accelerate the computationally taxing process of deep reinforcement learning (DRL) in numerical simulations for active control of bluff body flows at high Reynolds number ( $Re$ ) using transfer learning. We consider the canonical flow past a circular cylinder whose wake is controlled by two small rotating cylinders. We first pre-train the DRL agent using data from inexpensive simulations at low $Re$ , and subsequently we train the agent with small data from the simulation at high $Re$ (up to $Re=1.4\times 10^5$ ). We apply transfer learning (TL) to three different tasks, the results of which show that TL can greatly reduce the training episodes, while the control method selected by TL is more stable compared with training DRL from scratch. We analyse for the first time the wake flow at $Re=1.4\times 10^5$ in detail and discover that the hydrodynamic forces on the two rotating control cylinders are not symmetric.

An Early Detection and Classification of Alzheimer's Disease Framework Based on ResNet-50.

The objective of this study is to develop a more effective early detection system for Alzheimer's disease (AD) using a Deep Residual Network (ResNet) model by addressing the issue of convolutional layers in conventional Convolutional Neural Networks (CNN) and applying image preprocessing techniques. The proposed method involves using Contrast Limited Adaptive Histogram Equalizer (CLAHE) and Boosted Anisotropic Diffusion Filters (BADF) for equalization and noise removal and K-means clustering for segmentation. A ResNet-50 model with shortcut links between three residual layers is proposed to extract features more efficiently. ResNet-50 is preferred over other ResNet types due to its intermediate depth, striking a balance between computational efficiency and improved performance, making it a widely adopted and effective architecture for various computer vision tasks. While other ResNet variations may offer higher depths, they are more prone to overfitting and computational complexity, which can hinder their practical application. The proposed method is evaluated on a dataset of MRI scans of AD patients. The proposed method achieved high accuracy and minimum losses of 95% and 0.12, respectively. While some models showed better accuracy, they were prone to overfitting. In contrast, the suggested framework, based on the ResNet-50 model, demonstrated superior performance in terms of various performance metrics, providing a robust and reliable approach to Alzheimer's disease categorization. The proposed ResNet-50 model with shortcut links between three residual layers, combined with image preprocessing techniques, provides an effective early detection system for AD. The study demonstrates the potential of deep learning and image processing techniques in developing accurate and efficient diagnostic tools for AD. The proposed method improves the existing approaches to AD classification and provides a promising framework for future research in this area.

Cow detection and tracking system utilizing multi-feature tracking algorithm

In modern cattle farm management systems, video-based monitoring has become important in analyzing the high-level behavior of cattle for monitoring their health and predicting calving for providing timely assistance. Conventionally, sensors have been used for detecting and tracking their activities. As the body-attached sensors cause stress, video cameras can be used as an alternative. However, identifying and tracking individual cattle can be difficult, especially for black and brown varieties that are so similar in appearance. Therefore, we propose a new method of using video cameras for recognizing cattle and tracking their whereabouts. In our approach, we applied a combination of deep learning and image processing techniques to build a robust system. The proposed system processes images in separate stages, namely data pre-processing, cow detection, and cow tracking. Cow detection is performed using a popular instance segmentation network. In the cow tracking stage, for successively associating each cow with the corresponding one in the next frame, we employed the following three features: cow location, appearance features, as well as recent features of the cow region. In doing so, we simply exploited the distance between two gravity center locations of the cow regions. As color and texture suitably define the appearance of an object, we analyze the most appropriate color space to extract color moment features and use a Co-occurrence Matrix (CM) for textural representation. Deep features are extracted from recent cow images using a Convolutional Neural Network (CNN features) and are also jointly applied in the tracking process to boost system performance. We also proposed a robust Multiple Object Tracking (MOT) algorithm for cow tracking by employing multiple features from the cow region. The experimental results proved that our proposed system could handle the problems of MOT and produce reliable performance.

Early Detection of Lameness in Cattle with Image Processing Techniques

Lameness in dairy cattle has a negative effect on fertility, milk yield and various behaviors. Therefore, lameness in cattle causes significant economic losses in countries. In our article, it is aimed to detect lameness in cattle early with image processing techniques. Deep learning and image processing techniques were used in the article. In the article, YOLOv5 algorithm is used for object detection and Shufflenetv2k30 algorithm is used as image processing technology. Within the scope of the article, the images were subjected to a preprocessing (data augmentation) and then the cattle in the selected photos were identified by our trained deep learning model. The detected cattle were tagged and then the posture estimation of these tagged cattle was made. The angles between the joints of the cattle were found on the cattle whose posture pose was estimated. In the performance analysis, training was started with the weights of the Pre-training yolov5l model and the best weight output of the 200 epoch trained model was 75%. The best weight output of the model trained from zero to 400 epochs without using any model weights was 63%. Pre-training was started with the weights of the shufflenetv2k30 model and the weight output of the model trained for 400 epochs was 71%. This article will contribute to the studies to be done in the academic field and will create important data for the studies to be done in the livestock sector.

Intelligent Detection of Rebar Size and Position Using Improved DeeplabV3+

For the development of reinforced concrete structures and infrastructure construction, traditional rebar checking and acceptance methods have shortcomings in terms of efficiency. The use of digital image processing technology cannot easily identify a rebar configuration with complex and diverse backgrounds. To solve this problem, an inspection method combining deep learning and digital image processing techniques is proposed using an improved DeeplabV3+ model to identify reinforcing bars, with the identification results subjected to digital image processing operations to obtain the size information of the reinforcing bar. The proposed method was validated through a field test. The results of the experiment indicated that the proposed model is more accurate than other models, with a mean Intersection over Union (mIoU), precision, recall, and F1 score reaching 94.62%, 97.42%, 96.95%, and 97.18%, respectively. Moreover, the accuracy of the dimension estimations for the test reinforcements met the engineering acceptance standards.

MCSTransWnet: A new deep learning process for postoperative corneal topography prediction based on raw multimodal data from the Pentacam HR system

This work provides a new multimodal fusion generative adversarial net (GAN) model, Multiple Conditions Transform W-net (MCSTransWnet), which primarily uses femtosecond laser arcuate keratotomy surgical parameters and preoperative corneal topography to predict postoperative corneal topography in astigmatism-corrected patients. The MCSTransWnet model comprises a generator and a discriminator, and the generator is composed of two sub-generators. The first sub-generator extracts features using the U-net model, vision transform (ViT) and a multi-parameter conditional module branch. ​The second sub-generator uses a U-net network for further image denoising. The discriminator uses the pixel discriminator in Pix2Pix. Currently, most GAN models are convolutional neural networks; however, due to their feature extraction locality, it is difficult to comprehend the relationships among global features. Thus, we added a vision Transform network as the model branch to extract the global features. It is normally difficult to train the transformer, and image noise and geometric information loss are likely. Hence, we adopted the standard U-net fusion scheme and transform network as the generator, so that global features, local features, and rich image details could be obtained simultaneously. Our experimental results clearly demonstrate that MCSTransWnet successfully predicts postoperative corneal topographies (structural similarity ​= ​0.765, peak signal-to-noise ratio ​= ​16.012, and Fréchet inception distance ​= ​9.264). Using this technique to obtain the rough shape of the postoperative corneal topography in advance gives clinicians more references and guides changes to surgical planning and improves the success rate of surgery.

Cusp bifurcation in a metastatic regulatory network

Understanding the potential for cancers to metastasize is still relatively unknown. While many predictive methods may use deep learning or stochastic processes, we highlight a long standing mathematical concept that may be useful for modeling metastatic breast cancer systems. Ordinary differential equations (ODEs) can model cell state transitions by considering the pertinent environmental variables as well as the paths systems take over time. Bifurcation theory is a branch of dynamical systems which studies changes in the behavior of an ODE system while one or more parameters are varied. Many studies have applied concepts in one-parameter bifurcation theory to model biological network dynamics, and cell division. However, studies of two-parameter bifurcations are much more rare. Two-parameter bifurcations have not been studied in metastatic systems. Here we show how a specific two-parameter bifurcation phenomenon called a cusp bifurcation separates two qualitatively different metastatic cell state transitions modalities and propose a new perspective on defining such transitions based on mathematical theory. We hope the observations and verification methods detailed here may help in the understanding of metastatic potential from a basic biological perspective and in clinical settings.

Deep Learning Approach for Age-related Macular Degeneration Detection Using Retinal Images: Efficacy Evaluation of Different Deep Learning Models

Age-related macular degeneration (AMD) is a typical fundus disease that affects the central vision of elderly people. It causes difficulties in everyday activities such as reading and recognizing faces. AMD can progress slowly or rapidly, and it leads to severe vision loss if left untreated. Therefore, early detection and diagnosis of AMD are crucial to prevent or delay vision impairment in the elderly. To handle this requirement, researchers are exploring deep learning-based models as an AI tool to assist ophthalmologist in AMD diagnosis. However, conducting an appropriate deep learning model for the AMD classification is challenging and cost-intensive. This research aims to evaluate the efficacy of various deep learning models for obtaining the best performance results when identifying AMD disease using retinal images. To meet this objective, the retinal images from the Department of Ophthalmology, the King Chulalongkorn Memorial Hospital, Thailand were collected for transfer learning and other publicly available datasets for testing. Then, seven deep learning models VGG19, Xception, DenseNet201, EfficientNetB7, InceptionV3, NASNetLarge, and ResNet152V2 were chosen to training for the 2-labels (Normal vs. AMD) and the 3-labels (Normal vs. Dry AMD vs. Wet AMD) classifications. From the experimental results, the DenseNet201 model with Dense block in its structure showed the best efficacy in both 2-labels and 3-labels AMD classifications since its performance always include in the Top-3 models accuracy and generalization performance measured by total accuracy and total F1-Score, respectively. Furthermore, the accuracy performance of deep learning models in Top-3 are comparable with the performance of retinal specialist. These results contribute consolidated knowledge to the process of implementation effective deep learning as production that detects AMD automatically in the clinical and enhance the quality of healthcare service.

Deep Learning and image processing based ATM security and identifying the face

Deep Learning and image processing based ATM security and identifying the face

Research trends in the control of hate speech on social media for the 2016–2022 time frame

The growth in the number of social media users has resulted in a corresponding rise in the spread of hate speech on these platforms, leading to a growing, but little studied, problem. The bibliometric study aimed to examine the research trend and identify the most productive authors, the most active institutions, the leading countries and the most employed virtual hate speech control mechanisms by analyzing 576 relevant publications from the Scopus database published between 2016-2022. The findings showed an increase in publication and India as a leading country/region in research on virtual hate speech control mechanisms. Deep learning and natural language processing systems were identified as the most commonly used control mechanisms. Based on the results, it is recommended that future researchers focus on multidisciplinary collaboration and valid mechanisms for different languages. This paper provides a general overview of the current state of research in this field and serves as a guide for authors and institutions in their research and collaboration strategies.

Skull Stripping Based on the Segmentation Models

Skull image separation is one of the initial procedures used to detect brain abnormalities. In an MRI image of the brain, this process involves distinguishing the tissue that makes up the brain from the tissue that does not make up the brain. Even for experienced radiologists, separating the brain from the skull is a difficult task, and the accuracy of the results can vary quite a little from one individual to the next. Therefore, skull stripping in brain magnetic resonance volume has become increasingly popular due to the requirement for a dependable, accurate, and thorough method for processing brain datasets. Furthermore, skull stripping must be performed accurately for neuroimaging diagnostic systems since neither non-brain tissues nor the removal of brain sections can be addressed in the subsequent steps, resulting in an unfixed mistake during further analysis. Therefore, accurate skull stripping is necessary for neuroimaging diagnostic systems. This paper proposes a system based on deep learning and Image processing, an innovative method for converting a pre-trained model into another type of pre-trainer using pre-processing operations and the CLAHE filter as a critical phase. The global IBSR data set was used as a test and training set. For the system's efficacy, work was performed based on the principle of three dimensions and three sections of MR images and two-dimensional images, and the results were 99.9% accurate.

A deep reinforcement learning strategy for autonomous robot flocking

<p>Social behaviors in animals such as bees, ants, and birds have shown high levels of intelligence from a multi-agent system perspective. They present viable solutions to real-world problems, particularly in navigating constrained environments with simple robotic platforms. Among these behaviors is swarm flocking, which has been extensively studied for this purpose. Flocking algorithms have been developed from basic behavioral rules, which often require parameter tuning for specific applications. However, the lack of a general formulation for tuning has made these strategies difficult to implement in various real conditions, and even to replicate laboratory behaviors. In this paper, we propose a flocking scheme for small autonomous robots that can self-learn in dynamic environments, derived from a deep reinforcement learning process. Our approach achieves flocking independently of population size and environmental characteristics, with minimal external intervention. Our multi-agent system model considers each agent’s action as a linear function dynamically adjusting the motion according to interactions with other agents and the environment. Our strategy is an important contribution toward real-world flocking implementation. We demonstrate that our approach allows for autonomous flocking in the system without requiring specific parameter tuning, making it ideal for applications where there is a need for simple robotic platforms to navigate in dynamic environments.</p>