Smallest Error Value Research Articles

Decubitus pump health device implementation with pressure sensor based on Atmega 328

The decubitus, also known as pressure ulcer, is a wound that occurs due to pressure mainly on the protruding bones due to the patient is lying in bed on long time. This research purpose is to help prevent decubitus sores for immobility patients who are more bedridden during treatment at the hospital. This research module is a special mattress with an automatic pump which is equipped with an air pressure monitoring system on the mattress. The MPX5700 sensor measures mattress pressure. On the module testing, it can be observing that the largest error value on the low setting value measurement that is 2.25 % while the smallest error value on high setting measurement that is 1.84 %. Based on the results of tests and supported by the research, it can be concluded that the decubitus pump module can run well because it is still at the error tolerance threshold at ± 5 %.

A Modification of Infant Warmer with Monitoring of Oxygen Saturation, Heart Rate and Skin Temperature

Full term and premature babies have decreased body temperature by 0.1 ° C-0.3 ° C per minute, infant warmer was developed to provide the effect of heat on the baby as the temperature in the mother's uterus. The purpose of this study is to modify the infant warmer tool by adding parameters SpO2, BPM, and skin sensors. The contribution of this research is that the SpO2, BPM, and skin sensor monitoring systems are very helpful in the therapeutic and healing process using infant warmers. The use of the right sensor and has good quality will produce results and a small error value so it does not exceed the maximum value that is tolerated. The MAX 30100 sensor can detect oxygen saturation in the blood (SpO2) and heart rate (BPM), the DS18B20 temperature sensor can detect temperatures on the surface of the skin or body. Based on the results of measurements on 3 different respondents with 6 times of data retrieval and compared with standard tools the largest error value obtained from SpO2 is 0.68% and the smallest error value is 0.17%, while the largest error value from BPM is 0.76 bpm and the smallest error value is 0.42 bpm, the value of skin temperature error obtained is 0.28% obtained from the measurement of 1 respondent with 6 times of data retrieval and compared with standard tools. The tool is said to be suitable for use because the maximum limit in SpO2 error tolerance is 2% and BPM is 5%. The results of this research can be implemented on infant warmers to improve the monitoring of physiological conditions in patients.

Machine learning and registration for automatic seed localization in 3D US images for prostate brachytherapy.

New radiation therapy protocols, in particular adaptive, focal or boost brachytherapy treatments, require determining precisely the position and orientation of the implanted radioactive seeds from real-time ultrasound (US) images. This is necessary to compare them to the planned one and to adjust automatically the dosimetric plan accordingly for next seeds implantations. The image modality, the small size of the seeds, and the artifacts they produce make it a very challenging problem. The objective of the presented work is to setup and to evaluate a robust and automatic method for seed localization in three-dimensional (3D) US images. The presented method is based on a prelocalization of the needles through which the seeds are injected in the prostate. This prelocalization allows focusing the search on a region of interest (ROI) around the needle tip. Seeds localization starts by binarizing the ROI and removing false positives using, respectively, a Bayesian classifier and a support vector machine (SVM). This is followed by a registration stage using first an iterative closest point (ICP) for localizing the connected set of seeds (named strand) inserted through a needle, and secondly refining each seed position using sum of squared differences (SSD) as a similarity criterion. ICP registers a geometric model of the strand to the candidate voxels while SSD compares an appearance model of a single seed to a subset of the image. The method was evaluated both for 3D images of an Agar-agar phantom and a dataset of clinical 3D images. It was tested on stranded and on loose seeds. Results on phantom and clinical images were compared with a manual localization giving mean errors of 1.09±0.61mm on phantom image and 1.44±0.45mm on clinical images. On clinical images, the mean errors of individual seeds orientation was . The proposed algorithm for radioactive seed localization is robust, tested on different US images, accurate, giving small mean error values, and returns the five cylindrical seeds degrees of freedom.

Deep Learning-Aided Sensorless Control Approach for PV Converters in DC Nanogrids

In a microgrid, photovoltaic (PV) systems are broadly preferred with energy storage systems (ESSs) that form small-sized direct current (DC) microgrids. They are also termed local grids i.e., DC nanogrids, which feed the local consumers to some extent in the next decades. Therefore, ESSs enable the DC nanogrids more flexible and stable by preserving the intermittent nature of renewables. Yet still, feeding local consumers smoothly with PV-battery-based systems is exceedingly a considerable theme. In this context, proper control of power electronics converters as the main carrier of the system is essential. Besides, the rise of PV applications challenges possible issues upon integrating the conventional grid. Emerging possible issues in stability, reliability, efficiency and the ways of dealing with them have been developing day by day. Thus, it is inevitable that innovative methods will be put into practice. To achieve this goal, the deep learning aided-sensorless control approach is adopted. To validate the proposed control method, the training phase is presented elaborately with the help of the experimental setup of a DC nanogrid. From the obtained results, it is concluded that the deep learning-based approach reaches very small error values, captures the system dynamics successfully, enables a flexible structure with tunable hyper-parameters, and allows the possibility to apply practically.

МОДЕЛЬ QSPR ДЛЯ ПРОГНОЗА ОКТАНОВЫХ ЧИСЕЛ УГЛЕВОДОРОДОВ РЯДА АЛКЕНОВ ПО ТОПОЛОГИЧЕСКИМ ХАРАКТЕРИСТИКАМ МОЛЕКУЛ

Relevance Modeling, optimization and design of processes in the technology of oil refining and organic synthesis, as well as the use of new chemical and technological processes involv ing hydrocarbon systems with the calculation of physical and chemical properties (РCS) of liquids and gases. The accuracy of calculations plays an observed role in the calculations of oil refining and organic synthesis processes, as well as approximate reactor processes, fractionation and heat exchange equipment. Despite the availability of РCS databases, the search for adequate initial data takes a long time at the stage of the entire development or design process, since the РCS of many hydrocarbons remains unknown. Aim of research To develop a quantitative Structure-Property Ratio (QSPR) model for octane hydrocarbons of a series of alkenes. Research methods To predict the octane numbers of normal and substituted alkenes — components of catalytic cracking, a nonlinear multivariate regression model Quantitative Structure-Property Relationship (QSPR) is proposed. The objects of the study were 30 hydrocarbons of a number of alkenes, selection in the basic and test samples, made randomly using computer data of physical and chemical properties. The model associates a set of descriptors with the octane numbers — the topological characteristics of their molecular graphs: the Wiener index, the Randich index and the magnitude of the quadratic dependence of bone structure, which affect the octane numbers and reflect the main structural and chemical factors, such as the length and branching of the carbon skeleton, and sensitive parameters. molecules. Results The adequacy of the models was confirmed by statistical data processing, so the coefficient of determining the models is 0.856. For the quality characteristics of the QSPR model, the multiple correlation coefficient r = 0.925 was calculated, which suggested a force relationship between the proposed topological characteristics of hydrocarbon molecules and their octane numbers. To assess the statistical stability of the model, a correlation correction was used. The maximum absolute and relative errors for octane number sampling tests are 4.0 units and 4.1 %, respectively. The statistical calculation, which makes it possible to judge the adequacy of the predicted indicators, their compliance with the reference data, is the standard regression error of 6.5 units. The small value of the standard error of the regression in comparison with the values of the dependence of the applicable adequacy of the proposed model. The model adequately uses the octane numbers of linear and branched alkenes and can be used to predict the octane numbers of gasoline components.

Position Report Enhancement Using Bayesian Estimator

In this article, a Bayesian estimator for a target position report is proposed. It is based on a maximum a priori algorithm, where the user's device knowledge about its location is used to deduce the prior probability density function. The algorithm does not require knowledge about the signal and noise levels, meaning, noninformative priors are used. It is well known that with a higher number of antenna elements in an array, narrower beams can be formed. In beamforming, task narrow beams are useful for serving many users simultaneously, whereas in direction-of-arrival (DOA) estimation, we are not interested in narrow beams as such; instead, estimation accuracy is important. So, reducing the number of antenna elements used for DOA estimation is beneficial from a system complexity point of view. In this article, signal source location report is used to enhance the estimated DOA, for the task, MAP estimator is developed. We will show that in the case of small array size and large array covariance matrix error values, the proposed estimator is the only one capable of improving the prior knowledge about the transmitter when comparing it with other popular algorithms, such as the maximum likelihood estimator and Root MUSIC. The algorithm could be used in a variety of different scenarios, but its advantages emerge in the case of complex signal propagation environments, such as urban canyons and large airports.

Perbandingan Eksperimental dan Simulasi Frekuensi Pribadi pada Struktur Spindel CNC

The Research was about the comparison between experiment and simulation of natural frequency in CNC spindle. CNC spindle vibration will reduce machine tool performance. It could lead to the damage of the machine tool. The spindle structure unbalances of machine tools will cause vibration when it is operated. In the CNC machine, the spindle shaft vibration should be minimum. Based on this point, the natural frequency testing on the spindle shaft structure was carried out. The experiments were conducted by employing oscilloscope which could provide the vibration data in the time domain. The data was converted into the frequency domain using FFT. Measurements were carried out on 7 times of testing. Every one time of testing, 10 data were taken at each testing points. The tests were conducted at 10 testing points. Therefore, the total data obtained were 700 test data. The test results were then compared with the results of simulation modeling in 10 vibrate modes using Solidwork software. After testing and simulations were compared, 4 personal frequency values were obtained in the test that uses a measuring instrument and 6 personal frequency values could not be read. These were because the accelerometer used could not read up to 0 Hz frequency. Natural frequency obtained from simulations and tests were expressed in the percentage of errors. The largest error value in the 9th vibration mode measurement with a natural frequency was 2117.96 Hz with an error of 0.32%. The smallest error value was 0.11% with a natural frequency of 2995.79 Hz.

PERAMALAN JUMLAH PENUMPANG PESAWAT DI BANDARA INTERNASIONAL AHMAD YANI DENGAN METODE HOLT WINTER’S EXPONENTIAL SMOOTHING DAN METODE EXPONENTIAL EMOOTHING EVENT BASED

Forecasting the number of airplane passengers can be a consideration for the airline at Ahmad Yani International Airport related with addition of extra flight. The number of airplane passengers can be influenced by certain seasonal or special events. The seasonal influences can be known through historical data patterns and if there is a seasonal pattern, the Holt Winter’s Exponential Smoothing method can be used. Exponential Smoothing Event Based (ESEB) forecasting method can be use to see the special events that effect the number of airplane passengers at Ahmad Yani International Airport. After compared, the Holt Winter’s Exponential Smoothing method is a better method of forecasting the number of airplane passengers at Ahmad Yani International Airport because it has a smaller error value, namely the MSE value and the MAPE value than the Exponential Smoothing Event Based (ESEB)method. The MAPE and MSE values be produced from the best method each of 5,644139% and 619,998,718 .Keywords : Airplane Passengers, Seasonal Pattern, Special Event, Exponential Smoothing Event Based , Holt Winter’s Exponential Smoothing.

Predicting cetane number in diesel fuels using FTIR spectroscopy and PLS regression

Cetane number (CN) is an important property which indicates the ignition quality of fuels and especially diesel oil. The usual method for CN determination is a most involving and risky task that requires specific devices. In this paper, Partial Least Square Regression (PLSR) was successfully used for the prediction of diesel cetane number based on Fourier Transform Infrared Spectroscopy (FTIR). The proposed model was characterized by a high correlation coefficient between real and predicted CN values (R2 = 0.99), with small prediction error values (RMSEC = 0.28 and RMSEP = 0.42) compared to previously published models developed using spectroscopic techniques, namely NIR and Raman spectroscopy Thus, the proposed approach that uses the FTIR spectroscopy for cetane number determination can be highly recommended as a clean, environment friendly, rapid and reliable solution for the prediction of this important quality parameter of diesel fuels.

Numerical analysis of fractional viscoelastic column based on shifted Chebyshev wavelet function

An innovative numerical procedure for solving the viscoelastic column problem based on fractional rheological models, directly in the time domain, is investigated. Firstly, the governing equation is established according to the fractional constitutive relation. Secondly, the resulting equation is transformed into algebraic equation and solved by using the shifted Chebyshev wavelet function. Furthermore, the convergence analysis and the retained numerical benchmarks are carried out to validate the performance of the proposed method. A small value of the absolute error between numerical and accurate solution is obtained. Finally, the dynamic analysis of viscoelastic beam-column problems is investigated with different cross-section shape (circular and square) under various loading conditions (axial compressive force and harmonic load). The displacement, strain and stress of the viscoelastic column at different time and position are determined. The deformation and stress of the viscoelastic column of different materials under the same loading condition are compared. The results in the paper show the highly accuracy and efficiency of the proposed numerical algorithm in the dynamical stability analysis of the viscoelastic column.

Electroencephalogram (EEG) Signal Classification Using Artificial Neural Network to Control Electric Artificial Hand Movement

All due to the complex nature of the electroencephalography (EEG) signal, it is a challenge to be able to use it as the driver of an electric artificial hand. By using EEG signal, the command for artificial hand movements becomes more intuitive and natural. This study aims to classify EEG signals to serve as electronic hand control. Classification is conducted using artificial neural networks (ANN), in which EEG signal datasets are obtained from a commercial brain computer interface (BCI). The ANN model obtained is expected to be able to determine that the EEG signal is one of the five EEG signals generated from five predetermined hand movements. This study proposes feature extraction and processing that is very simple but performs well, indicated by its small error value. The results show that ANN can classify five hand movements tested with an overall accuracy rate of 80%.

2D Shallow Water Model for Dam Break and Column Interactions

Dam break causes disastrous effects on the surrounding area, especially at the downstream, therefore, there is a need for accurate and timely predictions of dam break propagation to prevent both property damage and loss of life. This study aimed to determine the movement of dam-break flow in the downstream area by solving the Shallow Water Equations (SWE) or Saint Venant Equations which are based on the conservation of mass and momentum derived from Navier Stokes equation. The model was generated using a finite difference scheme which is the most common and simplest method for dam-break modeling while Forward Time Central Space (FTCS) numerical scheme was applied to simulate two-dimensional SWE. Moreover, the accuracy of the numerical model was checked by comparing its results with the analytic results of one-dimensional cases and a relatively small value of error was found in comparison to the analytic models as indicated with the RMSE values close to 0. The numerical to the two-dimensional models were also compared to a simple dam break in a flume and dam break with column interactions and the wave propagation in both cases was observed to become very close at a certain time. The model, however, used numerical filter (Hansen) to reduce the oscillations or numerical instability. The simulation and analysis, therefore, showed the ability of the numerical scheme of FTCS to resolve both cases of the simple dam break and dam break with column interactions in the Two-dimensional Shallow Water.

Comparative analysis of rapid quality evaluation of Salvia miltiorrhiza (Danshen) with Fourier transform near-infrared spectrometer and portable near-infrared spectrometer

The root of Salvia miltiorrhiza, has wide applications in clinical practice such as the treatment of cardiovascular and cerebrovascular diseases. Due to the diversity of geographical origins and difference in quality, rapid and nondestructive methods such as near infrared spectroscopy (NIRS) technique for quality analysis are highly needed. Besides, because of its portability, the portable near-infrared spectrometer is suitable for the on-site detection of Chinese medicinal materials. It is not clear how far its performance differs from conventional analytical-grade NIR spectrometers. Therefore, in this study, a Fourier transform near-infrared spectrometer and a portable near-infrared spectrometer were both employed and compared in developing a rapid and nondestructive method for discrimination of geographical origins, determination of four active components (ACs, including tanshinone II A, cryptotanshinone, tanshinone I and salvianolic acid B) and evaluation of antioxidant activity (AA) in vitro. The geographical origins were discriminated by using partial least squares-discriminant analysis (PLS-DA) and the satisfactory discrimination results can be obtained by both instruments. PLS regression was applied to establish the quantification models for ACs and AA by NIRS. The correlation coefficients in the prediction set (Rp2) of Fourier transform near-infrared spectroscopy (FT-NIRS) for all ACs were above 0.93 with small prediction error and residual prediction deviation (RPD) values all greater than 3. While, the micro near-infrared spectroscopy (MicroNIRS) models performed slightly worse: the Rp2 values for most components can reach 0.92 with RPD values higher than 3, which was still enough for prediction, but for cryptotanshinone, the prediction error was larger and it may be difficult to guarantee accurate prediction results with limited samples. In the prediction of AA, since the reference values of AA were influenced by reaction contingency and experimental manipulation, several variable selection methods were then used to improve the evaluation of AA, and the RPD value of FT-NIRS can be increased from 3.3 to 4.1, while model of MicroNIRS became worse after variable selection due to the limited number of spectral variables and low resolution. According to the comparative analysis of the two instruments, this study could provide a useful reference and basis for instrument performance and accessible accuracy when performing rapid quality analysis of S. miltiorrhiza by NIRS.

Simulation-based optimization of plastic injection molding parameters for mini centrifugal pump body using response surface methodology

Control parameters of the injection molding and fabrication process are essential to the success of any micro-molding product fabrication. The goal of the study is to simulate and optimise the parameter for the plastic mini micro centrifugal pump body by using Response Surface Methodology. The material for this micro part is Polyphenylene Sulphide (PPS). This study is also investigating the impact of design parameters and processing parameters on the quality of a plastic moulded part to obtain optimal response and meets the requirement specification. The control factors chosen for this study are melting temperature, cooling time and injection pressure. Based on the complex shape of the body part of mini micro pump, volumetric shrinkage and warpage are selected as the essential quality that needs to be controlled. The selected processing parameter are melting temperature, cooling time and injection pressure. As a result, the optimum values suggested by the software are melt temperature of 315.98°C, cooling time of 362s and injection pressure of 62MPa. With small differences error value between solution and simulation, 0.27% for volumetric shrinkage and 0.19% for warpage, the result was acceptable.

FUTURE NATURAL GAS PRICE FORECASTING MODEL AND ITS POLICY IMPLICATION

Future natural gas (FNG) price is a collected data over the years and is a volatile movement in the market. In other words, FNG price is categorised as a time series data with volatility in both variance and mean, as well as most likely in some cases having heteroscedasticity problem. To come up with an estimated prediction model, some analysis tools, such as autoregressive integrated moving average (ARIMA) and generalised autoregressive conditional heteroscedasticity (GARCH), are introduced to find the best-fitted model having the smallest error value with high significance of probability value. This study aims to examine the best-fitted model that allows us to forecast FNG prices more accurately in the near future. There are 2842 observed data of daily FNG prices from 2009 to 2019 as the input of study objects. The finding suggests that the first measurement model of ARIMA (1,1,1) does not fit the model as having a non-significant probability value. Thus, it is required to check its heteroscedasticity by conducting an ARCH effect test. It is concluded that a data set has an effect of ARCH, so AR (p)–GARCH (p,q) model is then tested. AR (1)–GARCH (1,1) model is believed to be a best-fitted model having a significant P < 0.0001 with significantly small mean squared error and root mean squared error values, indicating that it has a very accurate prediction model. The forecasting model is to adjust the offered recommendation of policy for the government regarding the issue of high volatility of daily FNG prices in the future. We then offer a best-suited policy for some certain governments like Indonesia to give subsidy for targeted users in order to keep increasing their use of FNG that will expectedly affect their marketable product innovation and expansion, so economic growth in Indonesia is maintained.

PI Controller Untuk Mengatur Kecepatan Motor Induksi 1 Fasa

Induction motors are widely used in the industrial world, home-based businesses as well as in households. Currently in the process of making tofu an induction motor is used as a motor to drive soy blending blades. At this time the use of induction motors is still manually by requiring the operator to regulate the speed of the motor. To reduce operator work, it is necessary to apply PI control as a motor speed controller so that a constant motor rotation is obtained. 1 phase induction motor can be adjusted with variable speed drive (VSD) 0.75KW 1 phase. Blending blade drive uses 0.5HP 1 phase induction motor. In the application of PI control requires some hardware namely Arduino Uno as a minimum system that gives PWM circuit input commands. And the speed sensor as a motor blending speed reader. PI tuning values obtained from the application of the Ziegerl-Nichols I method with the best Kp and Ki tuning values are 1.35 and 0.02673. This research was conducted with 3 speed variables namely 1400 rpm, 1300 and 1200 rpm. From the application of Kp and Ki tuning values, the smallest error value is 4.08% at 1400 rpm with the system response time peak (tp) 5s, rise time (tr) 3s faster, delay time (td) 3s, and settling time (ts) 9s , and a maximum overshoot of 9.8%.

Forecasting the Number of Outpatient Patient Visits Using the ARIMA, SES And Holt-Winters Methods at XYZ Community Health Center

The purpose of the research is to find out the visiting patient at the community health center. This research is Forecasting used the ARIMA method, Single Exponential Smoothing, and Holt-Winters, which are very suitable for processing data that is time series as in outpatient visits. The data from outpatients is patient visits for five years from January 2014 to December 2018 where outpatient data were taken from the total number of outpatient visits from visit data categories: General, Clinic, BPJS, Non-BPJS (SKM ), Non-BPJS (Gakinda) List of new patients and the list of patients who were then predicted the level of patient visits during the next two years, namely from January 2019 to December 2020. This study compared the best method among the three-time series methods. The forecast compared with actual data to see accuracy and find out which Forecast is best. The final results show that the MAPE value of the ARIMA method for patient visit data is worth 22.55%, the Single Exponential Smoothing method is worth 9.74%, and the Holt-Winters method is worth 7.90%. It can be said that the smallest error value is Holt-Winters from patient visit data with MAPE 7.90%, which is said to be an excellent Forecasting category by producing a total value of Forecast = 53894.2 with an average monthly = 2245.59 for Forecasting for the next two years. This monthly average result is used as a reference for the number of visitors who come for each month around 2245 people after that the last step taken is to make a strategic design using the SWOT analysis technique combined with prediction results using the Holt-Winters method in getting a conclusion that can make documents which is only around 2245 documents/person, especially outpatient data for problem solutions that occur at public health centers.

Feature-Guided Nonrigid 3-D Point Set Registration Framework for Image-Guided Liver Surgery: From Isotropic Positional Noise to Anisotropic Positional Noise

Registration is an essential problem in image-guided surgery (IGS) since it brings different involved coordinate frames together. Nonrigid or deformable registration still faces many challenges, such as two point sets (PSs) are partially overlapped. To tackle the challenges in the nonrigid registration, we introduce a new two-step point-based registration pipeline that includes two steps. In the first step, the rigid transformation between the two spaces is recovered where the orientation vectors are adopted. In the second step, built on the nonrigid coherent point drift (CPD) approach, the anisotropic positional noise is also assumed. Registration results on the human liver verify the proposed approach' great improvements over the other methods. First, the rotation and translation are recovered with smaller error values than the existing methods. Second, our registration method's performance is much more robust to the partial overlapping between two PSs. Third, the two-step registration framework achieves the best performances in most test cases when there is a localization error in acquiring the intraoperative data. Note to Practitioners-A novel registration approach is presented for image-guided liver surgery (LGLS). Compared with existing nonrigid registration methods, two significant changes (or improvements) exist in the proposed registration framework: 1) the normal vectors are extracted and utilized in the rigid registration step and 2) the anisotropic positional uncertainties are considered. In both steps, the registration problems are formulated as a maximum likelihood (ML) problems and dealt with the expectation-maximization (EM) technique. In both steps, the matrix form of the updated positional covariance is provided and can speed up the computational process. The readers are reminded that with extra information and a more general positional error assumption, our approach demonstrates improved performances in the case of partial-to-full alignment.

FORECASTING OF UNMET NEEDS PERCENTAGE IN EAST JAVA PROVINCE USING AUTOREGRESSIVE INTEGRATED MOVING AVERAGE (ARIMA) METHOD

ARIMA uses present and past values as the dependent variable. The accuracy of the ARIMA forecasting method results is good to be used to obtain short-term forecasts. Compared to other time series methods, the advantage of ARIMA method is that it can be used in the percentage of unmet needs data in East Java Province since ARIMA method does not require any specific data motives. Unmet need is a group of women who do not want to have any more children or want to minimize their pregnancy but refuse to use contraception to prevent pregnancy. This study aims to determine the percentage of unmet needs in East Java Province in the future. This study will analyze the value of forecasting and determine the best model for ARIMA. The data used is the monthly data of unmet needs percentage of East Java Province starting from January 2014 to April 2019 (64 data plots). The results showed that the percentage of the number of unmet needs in East Java Province can be predicted using ARIMA model (12,1,0) without constant. The model is based on ARIMA (12,1,0) diagnostic test without constant meeting all the test requirements. The results of forecasting held a MAPE value of 2.369% and MAE of 0.26%. Based on MAPE and MAE, the model has a very good forecasting ability with a fairly small error value. Forecasting results indicated fluctuations in unmet needs data, where from December 2019 to February 2020 there was an increase in number of unmet needs in East Java Province. In the interim, starting in March 2020, the data needs in East Java Province tend to be constant at a higher position than the previous increase.

JSNet: A simulation network of JPEG lossy compression and restoration for robust image watermarking against JPEG attack

Deep learning-based watermarking methods have achieved a better performance in capacity and invisibility than some traditional methods. However, their robustness against JPEG lossy compression attack is still to be improved. To enhance the robustness and construct an end-to-end method, it is urgent to simulate the JPEG lossy compression by a neural network and then introduce it into the deep learning-based watermarking methods. In this paper, a JPEG simulation network JSNet is proposed to reappear the whole procedure of the JPEG lossy compression and restoration except entropy encoding as realistically as possible. The steps of sampling, DCT, and quantization are modeled by the max-pooling layer, convolution layer, and 3D noise-mask, respectively. The proposed JSNet can simulate JPEG lossy compression with any quality factors. To verify the proposed JSNet in improving the robustness against JPEG compression attack, a CNN-based robust watermarking network (CRWNet) is proposed as an application example. The end-to-end CRWNet contains three subnetworks, i.e., embedding subnetwork, JSNet, and extraction subnetwork. Here. the JSNet is regarded as an attack module in the pipeline. Experimental results on two publicly available datasets (ImageNet, BossBase) demonstrate that: (a) the proposed JSNet can well simulate JPEG lossy compression under any quality factors with small Root mean square error (RMSE) values; (b) the proposed CRWNet considering JSNet has achieved an average 30.6 percent advantage over the basic model without consideration of JSNet.