Step Size Research Articles

R-linear and strong convergence of Tseng methods for solving certain split and non-split problems

In the recent past, authors have fondly assumed the co-coercive condition for solving certain class of split inverse problems. Recently, some attempts have been made to relax this very strict condition, however, only few results exists in literature in such case. In this paper, we present some new Mann-type Tseng methods for solving split monotone variational inclusion and common fixed point problems for a finite collection of quasi-pseudocontractive operators. We prove strong convergence and R − linear convergence rate results of our methods, while the co-coerciveness property is dispensed with. Our methods incorporate the relaxation, inertial extrapolation and self-adaptive step size techniques to achieve better performance and faster convergence. With the aid of relevant numerical experiments, we showcase the performance profile of our methods in comparison with recent methods in literature.

The Influence of YOLOv5 Hyperparameters for Construction Details Detection

Computer vision has become a fundamental area of interest in recent decades. Each area has unique data which object detection methods can analyse. However, it is important to find the most suitable parameters for the model that detects different object groups. In this research has been investigated the influence of pre-trained YOLOv5 (nano (n), small (s), medium (m), large (l), extralarge (x)) models, hyperparameters (learning rate, momentum, and weight decay) and different image augmentation (hsv_h, degrees, translate, flipud, mosaic, mixup, shear, perspective) efficiency for similar construction details detection. A newly collected dataset with twenty-two labelled categories of construction details was prepared. A total of 270 models were trained and evaluated. Every model was evaluated with 3,300 test images which backgrounds were mixed, neutral, and white backgrounds. The most accurate model was YOLOv5l with learning rate – 0.001, momentum – 0.950 and weight decay – 0.0001. This model achieved – 0.5015 (50.15%) accuracy.

Nonsmooth Reduced Interface Models and Their Use in Co-Simulation of Mechanical Systems

Abstract In a co-simulation setup, the entire system is decomposed into a collection of individual subsystems that are interfaced together, with each subsystem being modeled and integrated separately according to its own requirements. To maintain the interconnectivity and consolidation of the primary system, these subsystems must communicate with each other through the interface and transfer certain information at the end points of a defined time interval termed macro time step. Inside the macro time step, the evolution of the interface variables has to be approximated as information about them will only be available again at the end of the step. In real-time simulations, the size of the macro time step and the accuracy of the approximated interface variables are critical factors; if the interface variables are approximated accurately, the size of the macro time step can be kept large enough to provide interactive rates without loss of accuracy and stability. This work focuses on systems where unilateral contact interactions are important and proposes reduced interface model concepts for such nonsmooth systems. The use of the proposed reduced interface model (RIM) is demonstrated in co-simulation to provide model-based approximation of the interface variables. The advantages of the proposed method are demonstrated through two representative case studies.

Dynamical behavior of a hepatitis B epidemic model and its NSFD scheme

AbstractHepatitis is inflammation of the liver, and one of its types, hepatitis B, is a contagious infection. Using mathematical models, the nature of the spread of the Hepatitis B virus can be predicted. In the present paper, a hepatitis B epidemic model with a Beddington–DeAngelis type incidence rate and a constant vaccination rate is considered. Some dynamical properties of this model, such as non-negativity, boundedness character, the basic reproduction number $$\mathcal {R}_0$$ R 0 , stability nature, and the bifurcation phenomenon, are investigated. By the Bendixson theorem, it is demonstrated that the disease-free equilibrium is globally asymptotically stable. It is shown that a transcritical bifurcation phenomenon occurs when $$\mathcal {R}_0=1$$ R 0 = 1 . It is concluded that the endemic equilibrium is globally asymptotically stable when $$\mathcal {R}_0>1$$ R 0 > 1 , by utilizing Dulac’s criteria. Also, a discrete system of difference equations is obtained by constructing a non-standard finite difference (NSFD) scheme for the continuous model. It is shown that the solutions of this discrete system are dynamically consistent for all finite step sizes. The theoretical results obtained are also supported and visualized by numerical simulations. These simulations also demonstrate that the NSFD scheme produces much more efficient results than the Euler or RK4 schemes, as shown in the theoretical results obtained.

A rehearsal framework for computational efficiency in online continual learning

AbstractIn the realm of online continual learning, models are expected to adapt to an ever-changing environment. One of the most persistent hurdles in this adaptation is the mitigation of a phenomenon called "Catastrophic Forgetting" (CF). This critical condition occurs when models trained on non-identically distributed data lose performance in previously learned tasks. Rehearsal methods, leveraging the ability to replay older samples, aim to address this challenge by incorporating a buffer of past training samples. However, the absence of known task boundaries complicates the adaptation of current CF mitigation methods. This paper proposes a method attuned to data stream characteristics and online model performance in a resource-constrained environment. The number of training iterations and learning rate emerges as crucial hyperparameters, impacting the efficacy and efficiency of online continual learning. Up to this point, we propose a combination of Experience Replay methodologies, a Drift Detector, and various training convergence policies, specially tailored for scenarios with unknown task boundaries. Experimental results demonstrate the effectiveness of our approach, maintaining or enhancing performance compared to baseline methods, while significantly improving computational efficiency.

Deep Learning Model for Cosmetic Gel Classification Based on a Short-Time Fourier Transform and Spectrogram.

Cosmetics and topical medications, such as gels, foams, creams, and lotions, are viscoelastic substances that are applied to the skin or mucous membranes. The human perception of these materials is complex and involves multiple sensory modalities. Traditional panel-based sensory evaluations have limitations due to individual differences in sensory receptors and factors such as age, race, and gender. Therefore, this study proposes a deep-learning-based method for systematically analyzing and effectively identifying the physical properties of cosmetic gels. Time-series friction signals generated by rubbing the gels were measured. These signals were preprocessed through short-time Fourier transform (STFT) and continuous wavelet transform (CWT), respectively, and the frequency factors that change over time were distinguished and analyzed. The deep learning model employed a ResNet-based convolution neural network (CNN) structure with optimization achieved through a learning rate scheduler. The optimized STFT-based 2D CNN model outperforms the CWT-based 2D and 1D CNN models. The optimized STFT-based 2D CNN model also demonstrated robustness and reliability through k-fold cross-validation. This study suggests the potential for an innovative approach to replace traditional expert panel evaluations and objectively assess the user experience of cosmetics.

An experimental investigation on the flow control of the partially stepped NACA0012 airfoil at low Reynolds numbers

An experimental study of flow control around a NACA0012 airfoil modified with fully and partially step geometry was conducted at Reynolds numbers of 6 × 104 and 1.2 × 105. Partially step geometries, the location of the step on the airfoil, and Reynolds number are varied to show their effect on aerodynamic performance and flow structures on the airfoil. Experimental results show that fully and partially step geometries are effective for flow control around the airfoil and aerodynamic performance enhancement. The maximum increase in lift coefficient is approximately 46%, and the stall angle is shifted about 1° by the SM4 model at Re of 6x104 while the step geometry is on the pressure side. For a Reynolds number of 1.2 × 105, the highest increase in the lift-to-drag ratio of the airfoil is observed about 17.1 at an angle of attack of 6° by the SM2 model when step geometry is placed on the pressure side of the airfoil. In the event of partially or fully stepped at the pressure side of the model, step flow structures, including the reattachment line, recirculation zone, and corner eddy, are obtained. However, even though the step geometry is at the suction surface of the model, the formation of a laminar to turbulent transition is observed. The overall results suggest that the partially stepped geometry length, Reynolds number, and step location side on the airfoil have an important role in flow control to improve aerodynamic performance for various angles of attack.

A Modified Perturb and Observe Algorithm in Photovoltaic Maximum Power Point Tracking System

Abstract: The power output of the photovoltaic (PV) array is nonlinear, and maximum power point tracking (MPPT) is required to boost the efficiency of solar energy generation. The traditional perturb and observe algorithm is frequently used, but it is difficult to comprehensively consider the tracking accuracy and response speed at the same time. Misjudgment occurs when the solar insolation changes drastically. In view of these shortcomings, this paper presents a variable step size threshold search algorithm. The modified algorithm can quickly track the maximum power point and restrain fluctuations near the maximum power point. Matlab is utilized to simulate and prove the effectiveness of this modified algorithm.

Monitoring of Chlorophyll Content of Potato in Northern Shaanxi Based on Different Spectral Parameters

Leaf chlorophyll content (LCC) is an important physiological index to evaluate the photosynthetic capacity and growth health of crops. In this investigation, the focus was placed on the chlorophyll content per unit of leaf area (LCCA) and the chlorophyll content per unit of fresh weight (LCCW) during the tuber formation phase of potatoes in Northern Shaanxi. Ground-based hyperspectral data were acquired for this purpose to formulate the vegetation index. The correlation coefficient method was used to obtain the “trilateral” parameters with the best correlation between potato LCCA and LCCW, empirical vegetation index, any two-band vegetation index constructed after 0–2 fractional differential transformation (step size 0.5), and the parameters with the highest correlation among the three spectral parameters, which were divided into four combinations as model inputs. The prediction models of potato LCCA and LCCW were constructed using the support vector machine (SVM), random forest (RF) and back propagation neural network (BPNN) algorithms. The results showed that, compared with the “trilateral” parameter and the empirical vegetation index, the spectral index constructed by the hyperspectral reflectance after differential transformation had a stronger correlation with potato LCCA and LCCW. Compared with no treatment, the correlation between spectral index and potato LCC and the prediction accuracy of the model showed a trend of decreasing after initial growth with the increase in differential order. The highest correlation index after 0–2 order differential treatment is DI, and the maximum correlation coefficients are 0.787, 0.798, 0.792, 0.788 and 0.756, respectively. The maximum value of the spectral index correlation coefficient after each order differential treatment corresponds to the red edge or near-infrared band. A comprehensive comparison shows that in the LCCA and LCCW estimation models, the RF model has the highest accuracy when combination 3 is used as the input variable. Therefore, it is more recommended to use the LCCA to estimate the chlorophyll content of crop leaves in the agricultural practices of the potato industry. The results of this study can enhance the scientific understanding and accurate simulation of potato canopy spectral information, provide a theoretical basis for the remote sensing inversion of crop growth, and promote the development of modern precision agriculture.

LSTM network optimization and task network construction based on heuristic algorithm

This work aims to advance the security management of complex networks to better align with evolving societal needs. The work employs the Ant Colony Optimization algorithm in conjunction with Long Short-Term Memory neural networks to reconstruct and optimize task networks derived from time series data. Additionally, a trend-based noise smoothing scheme is introduced to mitigate data noise effectively. The approach entails a thorough analysis of historical data, followed by applying trend-based noise smoothing, rendering the processed data more scientifically robust. Subsequently, the network reconstruction problem for time series data originating from one-dimensional dynamic equations is addressed using an algorithm based on the principles of Stochastic Gradient Descent (SGD). This algorithm decomposes time series data into smaller samples and yields optimal learning outcomes in conjunction with an adaptive learning rate SGD approach. Experimental results corroborate the remarkable fidelity of the weight matrix reconstructed by this algorithm to the true weight matrix. Moreover, the algorithm exhibits efficient convergence with increasing data volume, manifesting shorter time requirements per iteration while ensuring the attainment of optimal solutions. When the sample size remains constant, the algorithm’s execution time is directly proportional to the square of the number of nodes. Conversely, as the sample size scales, the SGD algorithm capitalizes on the availability of more information, resulting in improved learning outcomes. Notably, when the noise standard deviation is 0.01, models predicated on SGD and the Least-Squares Method (LSM) demonstrate reduced errors compared to instances with a noise standard deviation of 0.1, highlighting the sensitivity of LSM to noise. The proposed methodology offers valuable insights for advancing research in complex network studies.

Efficient reliability assessment of slope stability with insightful sensitivity analysis concerning the safety factor

The limit state functions (LSFs) of slope stability are nonlinear, implicit, or derived from numerical simulations, and additional surrogate models are usually used for its reliability problems. Alternatively, an efficient numerical procedure using a variant of the HLRF-BFGS algorithm is proposed to address this issue without additional surrogate models. This proposed procedure simultaneously conducts reliability assessment and sensitivity analysis concerning the safety factor by suggested calculations of partial derivatives using the finite difference method (FDM). Two highly nonlinear numerical examples are presented to validate its convergence and to investigate the step length of the FDM. Results show that the proposed procedure can get considerable accuracy while the HLRF algorithm cannot converge, and its required iterations are about 30% and 81% less than those of the limited step-size iteration algorithm. In addition, a step length not greater than 0.1 is suggested after investigations. Application to a classical two-layered slope is illustrated. Reliability results show that the proposed procedure can get the same accuracy as those of surrogate models after only 5 iterative steps. Sensitivity results show that the cohesion of the upper soil is the only sensitivity factor, and the most possible potential slide surface lies in the upper soil.

Multi unmanned aerial vehicle odor source location based on improved artificial fish swarm algorithm

Odor source location technology has important application value in environmental monitoring, safety emergency and search and rescue operations. For example, it can be used in post-disaster search and rescue, detection of hazardous gas leakage, and fire source detection. Existing odor source location methods have problems such as low search efficiency, inability to adapt to complex environments, and inaccurate odor source location. In this study, based on unmanned aerial vehicle technology and using swarm intelligence optimization algorithm, an improved artificial fish swarm algorithm (IAFSA) is proposed by combining curiosity in psychology on the basis of retaining the good optimization performance of the artificial fish swarm algorithm. The algorithm quantifies the curiosity of artificial fish searching high-concentration areas through a model, dynamically adjusts the artificial fish's field of vision and step length with the calculated curiosity factor, and avoids the oscillation phenomenon in the later stage of the algorithm. Simulation results show that the IAFSA has a higher success rate and smaller location error. Finally, odor source location experiments were carried out in an indoor physical environment, the feasibility of the odor source location method proposed in this study is verified in actual scenarios.

Recognition of Plastic Film in Terrain-Fragmented Areas Based on Drone Visible Light Images

In order to meet the growing demand for food and achieve food security development goals, contemporary agriculture increasingly depends on plastic coverings such as agricultural plastic films. The remote sensing-based identification of these plastic films has gradually become a necessary tool for agricultural production management and soil pollution prevention. Addressing the challenges posed by the complex terrain and fragmented land parcels in karst mountainous regions, as well as the frequent presence of cloudy and foggy weather conditions, the extraction efficacy of mulching films is compromised. This study utilized a DJI Mavic 2 Pro UAV to capture visible light images in an area with complex terrain features such as peaks and valleys. A plastic film sample dataset was constructed, and the U-Net deep learning model parameters integrated into ArcGIS Pro were continuously modified and optimized to achieve precise plastic film identification. The results are as follows: (1) Sample quantity significantly affects recognition performance. When the sample size is 800, the accuracy of plastic film extraction notably improves, with area accuracy reaching 91%, a patch quantity accuracy of 96.38%, and an IOU and F1-score of 85.89% and 94.20%, respectively, compared to the precision achieved with a sample size of 300; (2) Different learning rates, batch sizes, and iteration numbers have a certain impact on the training effectiveness of the U-Net model. The most suitable model parameters improved the training effectiveness, with the highest training accuracy achieved at a learning rate of 0.001, a batch size of 10, and 25 iterations; (3) Comparative experiments with the Support Vector Machine (SVM) model validate the suitability of U-Net model parameters and sample datasets for precise identification in rugged terrains with fragmented spatial distribution, particularly in karst mountainous regions. This underscores the applicability of the U-Net model in recognizing plastic film coverings in karst mountainous regions, offering valuable insights for agricultural environmental health assessment and green planting management in farmlands.

Effects of sedentary behaviour and long-term regular Tai Chi exercise on dynamic stability control during gait initiation in older women

Background: Sedentary behaviour has been associated with an increased risk of falls among older adults. Although gait initiation (GI) is a promising tool used to assess fall risk, it has yet to be quantitatively evaluated for dynamic stability in sedentary populations. Tai Chi exercise is believed to be effective in preventing falls in older adults, but its effect on GI stability has not been quantified. This study aims to compare the stability of GI in sedentary older individuals versus those who are long-term Tai Chi exercisers by using a quantitative approach.Methods: This study included 17 sedentary older women without exercise habits (age: 65.59 ± 3.66 years, average daily sitting time: 8.735 ± 1.847 h/day) and 19 older women who regularly engage in Tai Chi exercise (age: 65.58 ± 3.63 years, years of exercise: 9.84 ± 3.48 years). Every participant underwent five trials of self-paced GI walking tests. Eight cameras and four force plates were used to obtain kinematic and kinetic parameters. The trajectory of the centre of mass (CoM) and the position of the foot placement were recorded. The anterior–posterior (A-P) and medio–lateral (M-L) dynamic stability at the onset and end moments of the single-legged support was calculated using CoM and gait spatiotemporal parameters. The stepping dynamic stability and foot placement positions of both groups were compared.Results: The Tai Chi group had greater stability in the M-L directions at the swing leg’s toe-off moment and in the M-L and A-P directions at the heel-strike moment, as well as significantly larger step length, step width and step speed during locomotion than sedentary older women. However, the stability in the A-P directions at the swing leg’s toe-off moment and the foot inclination angle was not statistically different between the two groups.Conclusion: Long-term regular Tai Chi exercise can enhance the dynamic stability of GI in older women, and effectively improve their foot placement strategy during GI. The findings further confirm the negative effect of sedentary on the stability control of older women and the positive role of Tai Chi in enhancing their gait stability and reducing the risk of falls.

Efficient Design of Battery Thermal Management Systems for Improving Cooling Performance and Reducing Pressure Drop

The air-cooled system is one of the most widely used battery thermal management systems (BTMSs) for the safety of electric vehicles. In this study, an efficient design of air-cooled BTMSs is proposed for improving cooling performance and reducing pressure drop. Combining with a numerical calculation method, a strategy with a varied step length of adjustments (∆d) is developed to optimize the spacing distribution among battery cells for temperature uniformity improvement. The optimization results indicate that the developed strategy reduces the optimization time by about 50% compared with a strategy using identical ∆d values while maintaining good performance of the optimized system. Furthermore, the system’s pressure drop does not increase after the spacing optimization. Based on this characteristic, a structural design strategy is proposed to improve the cooling performance and reduce the pressure drop simultaneously. First, the appropriate flow pattern is arranged and the secondary outlet is added to reduce the pressure drop of the system. The results show that the BTMS with U-type flow combined with a secondary outlet against the original outlet can effectively reduce the pressure drop of the system. Subsequently, this BTMS is further improved using the developed cell spacing optimization strategy with varied ∆d values while the pressure drop is fixed. It is found that the final optimized BTMS achieves a battery temperature difference below 1 K for different inlet airflow rates, with the pressure drop being reduced by at least 45% compared with the BTMS before the optimization.

Multi-market P2P trading of cooling–heating-power-hydrogen integrated energy systems: An equilibrium-heuristic online prediction optimization approach

In this paper, an equilibrium-heuristic online prediction optimization approach is proposed for multi-market peer-to-peer (P2P) electricity–hydrogen trading of integrated energy systems (IESs) with uncertainties. First, the IES, consisting of a hydrogen energy storage subsystem and a combined cooling, heating, and power subsystem, is constructed in the distribution network to improve energy utilization and market efficiency. Then, a bi-level optimization model for IESs, participating in the P2P electricity–hydrogen energy trading, the real-time electricity, and the ancillary service markets, is proposed, in which the top-level model can be formulated as a P2P electricity–hydrogen trading pricing model through the social welfare maximization problem, and the lower-level model is used to maximize the IES operating profit. To effectively solve the bi-level model, the game equilibrium-based ADMM distributed algorithm is used to obtain the P2P trading volume and prices of the top-level model, and a new hybrid heuristic algorithm, called hybrid sand cat swarm optimization and improved honey badger algorithm (SCIHB), is proposed to solve the lower level model. SCIHB utilizes a hybrid mechanism and an adaptive learning rate parameter to balance local exploitation and global exploration. Furthermore, a new online learning probabilistic prediction method based on the hidden Markov model and wavelet transform is introduced to describe the uncertainties of wind power, electricity prices, and frequency regulation prices. Finally, case studies are conducted on an IEEE 33-bus test system, and the numerical results verify the effectiveness of the proposed P2P trading model and solution approach.

Reward Sensitivity and Noise Contribute to Negative Affective Bias: A Learning Signal Detection Theory Approach in Decision-Making

In patients with mood disorders, negative affective biases – systematically prioritising and interpreting information negatively – are common. A translational cognitive task testing this bias has shown that depressed patients have a reduced preference for a high reward under ambiguous decision-making conditions. The precise mechanisms underscoring this bias are, however, not yet understood. We therefore developed a set of measures to probe the underlying source of the behavioural bias by testing its relationship to a participant’s reward sensitivity, value sensitivity and reward learning rate. One-hundred-forty-eight participants completed three online behavioural tasks: the original ambiguous-cue decision-making task probing negative affective bias, a probabilistic reward learning task probing reward sensitivity and reward learning rate, and a gambling task probing value sensitivity. We modelled the learning task through a dynamic signal detection theory model and the gambling task through an expectation-maximisation prospect theory model. Reward sensitivity from the probabilistic reward task (β = 0.131, p = 0.024) and setting noise from the probabilistic reward task (β = –0.187, p = 0.028) both predicted the affective bias score in a logistic regression. Increased negative affective bias, at least on this specific task, may therefore be driven in part by a combination of reduced sensitivity to rewards and more variable responses.

Reward actively engages both implicit and explicit components in dual force field adaptation.

Motor learning occurs through multiple mechanisms, including unsupervised, supervised (error-based) and reinforcement (reward-based) learning. Although studies have shown that reward leads to an overall better motor adaptation, the specific processes by which reward influences adaptation are still unclear. Here, we examine how the presence of reward affects dual-adaptation to novel dynamics, and distinguish its influence on implicit and explicit learning. Participants adapted to two opposing force fields in an adaptation/de-adaptation/error-clamp paradigm, where five levels of reward (a score and a digital face) were provided as participants reduced their lateral error. Both reward and control (no reward provided) groups simultaneously adapted to both opposing force fields, exhibiting a similar final level of adaptation, which was primarily implicit. Triple-rate models fit to the adaptation process found higher learning rates in the fast and slow processes, and a slightly increased fast retention rate for the reward group. While differences in the slow learning rate were only driven by implicit learning, the large difference in the fast learning rate was mainly explicit. Overall, we confirm previous work showing that reward increases learning rates, extending this to dual-adaptation experiments, and demonstrating that reward influences both implicit and explicit adaptation. Specifically, we show that reward acts primarily explicitly on the fast learning rate and implicitly on the slow learning rates.

Rewilding landscapes with apex predators: cheetah (Acinonyx jubatus) movements reveal the importance of environmental and individual contexts

Rewilding landscapes through species or population restoration is an increasingly applied practice in biological conservation. There is expanding interest in wildlife release projects for apex predator population augmentation or reintroductions in historical ranges. Cheetahs (Acinonyx jubatus) are an IUCN Vulnerable-listed species with a declining global population facing major threats, which in southern Africa primarily include lethal persecution on livestock farms and bush encroachment transforming open habitats to woody areas. We used GPS radiocollars to monitor ten adult cheetahs from 2007 – 2018 in the Central Plateau of Namibia encompassing an area restored as an open savanna field (13.7 km2) located in a matrix of woodland savanna affected by bush encroachment. We generated a set of a priori hypotheses that tested the effects of various factors on cheetah movements indexed by step length. We compared cheetah movement metrics based on their history as wild, rehabilitated, and/or translocated individuals. Day/night activity, habitat type, and habitat edges were significant predictors of cheetah movement. Wild resident cheetahs displayed significantly longer steps than the other cheetah classes, possibly suggesting increased territorial behaviour in response to the presence of introduced cheetahs. Some cheetahs temporally segregated by moving extensively during daytime, but most individuals were primarily active during crepuscular periods. Small prey remained constant across time, whereas large prey declined over the study period. Cheetahs appeared to adjust behaviourally by increasing movements in years when large prey were scarce. Cheetahs appeared to be ecologically adaptable and behaviourally flexible in response to varying prey populations and when translocated to new environments, specifically at the interface between bush-encroached woodland and open savanna. Environmental settings and animal history need to be carefully considered in rewilding and ecosystem restoration, and monitoring of released and resident individuals, if present, is important to understand ecological dynamics at release sites.

Development of a Mobile-Based Application for Classifying Caladium Plants Using the CNN Algorithm

Caladium is a popular ornamental plant and has business potential. However, difficulties in recognizing the type of Caladium often occur because of the similarities in shape, pattern, and color of the leaves between the different kinds of Caladium. To overcome this problem, research will use machine learning with the Convolutional Neural Network (CNN) algorithm to build a mobile application that can accurately classify four types of Caladiums. The data set used is 1200 data with four classes; each class has 300 data. The best model is found with the parameter epoch 100, learning rate 0.001, and batch size 64. The model is then implemented in a mobile application with two menus, "Take a photo" and "Choose an image," which will display the classification output and confidence values of the four types of Caladiums. Testing with 30 test data per class achieves 0.975 accuracy on both menus. On the “Take a photo” menu, precision is 0.974, recall is 0.9725, and f1-score is 0.965. Meanwhile, on the “Choose an image” menu a precision and recall value is 0.975, and f1-score value of 0.97.