GoogLeNet Convolutional Neural Network Research Articles

Identification of Disc Cutter Wear via Operation Parameters Combined With Vibration Data: A Case Study

ABSTRACTThis paper proposed an approach to estimate disc cutter wear utilizing a combination of multiple operational parameters and vibration data collected during shield tunneling operations. The incorporation of vibration signals, notably those originating from acceleration sensors mounted on the back plate of the soil chamber, has markedly enhanced the accuracy of the model. Time‐frequency domain features were extracted through analysis methods such as Fast Fourier Transform (FFT), Short‐Time Fourier Transform (STFT), and Continuous Wavelet Transform (CWT). A predictive model utilizing vibration and shield operation parameters was developed using the XGBoost algorithm, and a deep GoogLeNet Convolutional Neural Network (CNN) was trained on time‐frequency graphs from the CWT. In addition, this study also investigated the impact of signal duration on wavelet image information and model accuracy. In the Huang‐Shang Intercity Railway Project, the approach effectively assessed disc cutter wear during tunneling operations and dynamically optimized the operational parameters of the shield tunnel machine through predictive analysis.

Deep Learning-Based Artificial Intelligence Can Differentiate Treatment-Resistant and Responsive Depression Cases with High Accuracy.

Background: Although there are many treatment options available for depression, a large portion of patients with depression are diagnosed with treatment-resistant depression (TRD), which is characterized by an inadequate response to antidepressant treatment. Identifying the TRD population is crucial in terms of saving time and resources in depression treatment. Recently several studies employed various methods on EEG datasets for automatic depression detection or treatment outcome prediction. However, no previous study has used the deep learning (DL) approach and EEG signals for detecting treatment resistance. Method: 77 patients with TRD, 43 patients with non-TRD, and 40 healthy controls were compared using GoogleNet convolutional neural network and DL on EEG data. Additionally, Class Activation Maps (CAMs) acquired from the TRD and non-TRD groups were used to obtain distinctive regions for classification. Results: GoogleNet classified the healthy controls and non-TRD group with 88.43%, the healthy controls and TRD subjects with 89.73%, and the TRD and non-TRD group with 90.05% accuracy. The external validation accuracy for the TRD-non-TRD classification was 73.33%. Finally, the CAM analysis revealed that the TRD group contained dominant features in class detection of deep learning architecture in almost all electrodes. Limitations: Our study is limited by the moderate sample size of clinical groups and the retrospective nature of the study. Conclusion: These findings suggest that EEG-based deep learning can be used to classify treatment resistance in depression and may in the future prove to be a useful tool in psychiatry practice to identify patients who need more vigorous intervention.

Utilizing deep learning via computer vision for agricultural production quality control: jackfruit growth stage identification

Jackfruit (Artocarpus heterophyllus), a tropical fruit renowned for its diverse culinary uses, necessitates identifying the optimal growth stage to ensure superior flavor and texture. This research investigates employing deep learning techniques, particularly convolutional neural networks (CNNs), for accurately detecting jackfruit growth stages. Despite the challenge posed by the nuanced visual differences among fruits at various maturity stages, a meticulously curated dataset of labeled jackfruit images was developed in collaboration with experts, utilizing the BBCH scale. This dataset facilitated training and evaluation. A modified version of the Places 365 GoogLeNet CNN model was proposed for classifying four distinct growth stages of jackfruit, compared with a state-of-the-art CNN model. The trained models demonstrated varying levels of accuracy in classification. Furthermore, the proposed CNN model was trained and tested using original and augmented images, achieving an impressive overall validation accuracy of 90%. These results underscore the efficacy of deep learning in automating the detection of growth stages, offering promising implications for quality control and decision-making in jackfruit production and distribution.

A real-time epilepsy seizure detection approach based on EEG using short-time Fourier transform and Google-Net convolutional neural network

Epilepsy is one of the most common brain disorders, and seizures of epilepsy have severe adverse effects on patients. Real-time epilepsy seizure detection using electroencephalography (EEG) signals is an important research area aimed at improving the diagnosis and treatment of epilepsy. This paper proposed a real-time approach based on EEG signal for detecting epilepsy seizures using the STFT and Google-net convolutional neural network (CNN). The CHB-MIT database was used to evaluate the performance, and received the results of 97.74 % in accuracy, 98.90 % in sensitivity, 1.94 % in false positive rate. Additionally, the proposed method was implemented in a real-time manner using the sliding window technique. The processing time of the proposed method just 0.02 s for every 2-s EEG episode and achieved average 9.85- second delay in each seizure onset.

Delamination Depth Detection in Composite Plates Using the Lamb Wave Technique Based on Convolutional Neural Networks.

Delamination represents one of the most significant and dangerous damages in composite plates. Recently, many papers have presented the capability of structural health monitoring (SHM) techniques for the investigation of structural delamination with various shapes and thickness depths. However, few studies have been conducted regarding the utilization of convolutional neural network (CNN) methods for automating the non-destructive testing (NDT) techniques database to identify the delamination size and depth. In this paper, an automated system qualified for distinguishing between pristine and damaged structures and classifying three classes of delamination with various depths is presented. This system includes a proposed CNN model and the Lamb wave technique. In this work, a unidirectional composite plate with three samples of delamination inserted at different depths was prepared for numerical and experimental investigations. In the numerical part, the guided wave propagation and interaction with three samples of delamination were studied to observe how the delamination depth can affect the scattered and trapped waves over the delamination region. This numerical study was validated experimentally using an efficient ultrasonic guided waves technique. This technique involved piezoelectric wafer active sensors (PWASs) and a scanning laser Doppler vibrometer (SLDV). Both numerical and experimental studies demonstrate that the delamination depth has a direct effect on the trapped waves' energy and distribution. Three different datasets were collected from the numerical and experimental studies, involving the numerical wavefield image dataset, experimental wavefield image dataset, and experimental wavenumber spectrum image dataset. These three datasets were used independently with the proposed CNN model to develop a system that can automatically classify four classes (pristine class and three different delamination classes). The results of all three datasets show the capability of the proposed CNN model for predicting the delamination depth with high accuracy. The proposed CNN model results of the three different datasets were validated using the GoogLeNet CNN. The results of both methods show an excellent agreement. The results proved the capability of the wavefield image and wavenumber spectrum datasets to be used as input data to the CNN for the detection of delamination depth.

ALL classification using neural ensemble and memetic deep feature optimization.

Acute lymphoblastic leukemia (ALL) is a fatal blood disorder characterized by the excessive proliferation of immature white blood cells, originating in the bone marrow. An effective prognosis and treatment of ALL calls for its accurate and timely detection. Deep convolutional neural networks (CNNs) have shown promising results in digital pathology. However, they face challenges in classifying different subtypes of leukemia due to their subtle morphological differences. This study proposes an improved pipeline for binary detection and sub-type classification of ALL from blood smear images. At first, a customized, 88 layers deep CNN is proposed and trained using transfer learning along with GoogleNet CNN to create an ensemble of features. Furthermore, this study models the feature selection problem as a combinatorial optimization problem and proposes a memetic version of binary whale optimization algorithm, incorporating Differential Evolution-based local search method to enhance the exploration and exploitation of feature search space. The proposed approach is validated using publicly available standard datasets containing peripheral blood smear images of various classes of ALL. An overall best average accuracy of 99.15% is achieved for binary classification of ALL with an 85% decrease in the feature vector, together with 99% precision and 98.8% sensitivity. For B-ALL sub-type classification, the best accuracy of 98.69% is attained with 98.7% precision and 99.57% specificity. The proposed methodology shows better performance metrics as compared with several existing studies.

AI-enabled crack-length estimation from acoustic emission signal signatures

Abstract This article addresses the classification of fatigue crack length using artificial intelligence (AI) applied to acoustic emission (AE) signals. The AE signals were collected during fatigue of two specimen types. One specimen type had a 1-mm hole for crack initiation. The other specimen type had a 150-micron wide slit of various lengths. Fatigue testing was performed under stress-intensity-factor control to moderate crack advancement. The slit specimen produced AE signals only from crack advancement at the slit tips whereas the 1-mm hole specimens produced AE signals both from crack tip advancement and crack rubbing or clapping. The AE signals were captured with a piezoelectric wafer active sensor (PWAS) array connected to MISTRAS instrumentation and AEwin software. The collected AE signals were preprocessed using time-of-flight filtering and denoising. Choi Williams transform converted time-domain AE-signals into spectrograms. To apply machine learning, the spectrogram images were used as input data for the training, validation, and testing of a GoogLeNet convolutional neural network (CNN). The CNN was trained to sort the AE signals into crack-length classes. CNN performance enhancements, including synthetic data generation and class balancing were developed. A three-class example with crack lengths of (i) 10-12 mm; (ii) 12-14 mm; and (iii) 14-16 mm is provided. Our AI approach was able to classify the AE signals into these three classes with 91% accuracy thus proving that the AE signals contain sufficient information for crack estimation using an AI-enabled approach.

Predicting heart disease using modified GoogLeNet convolutional neural network architecture based on the heart sound

Predicting heart disease using modified GoogLeNet convolutional neural network architecture based on the heart sound

Hybrid Feature-Learning-Based PSO-PCA Feature Engineering Approach for Blood Cancer Classification.

Acute lymphoblastic leukemia (ALL) is a lethal blood cancer that is characterized by an abnormal increased number of immature lymphocytes in the blood or bone marrow. For effective treatment of ALL, early assessment of the disease is essential. Manual examination of stained blood smear images is current practice for initially screening ALL. This practice is time-consuming and error-prone. In order to effectively diagnose ALL, numerous deep-learning-based computer vision systems have been developed for detecting ALL in blood peripheral images (BPIs). Such systems extract a huge number of image features and use them to perform the classification task. The extracted features may contain irrelevant or redundant features that could reduce classification accuracy and increase the running time of the classifier. Feature selection is considered an effective tool to mitigate the curse of the dimensionality problem and alleviate its corresponding shortcomings. One of the most effective dimensionality-reduction tools is principal component analysis (PCA), which maps input features into an orthogonal space and extracts the features that convey the highest variability from the data. Other feature selection approaches utilize evolutionary computation (EC) to search the feature space and localize optimal features. To profit from both feature selection approaches in improving the classification performance of ALL, in this study, a new hybrid deep-learning-based feature engineering approach is proposed. The introduced approach integrates the powerful capability of PCA and particle swarm optimization (PSO) approaches in selecting informative features from BPI mages with the power of pre-trained CNNs of feature extraction. Image features are first extracted through the feature-transfer capability of the GoogleNet convolutional neural network (CNN). PCA is utilized to generate a feature set of the principal components that covers 95% of the variability in the data. In parallel, bio-inspired particle swarm optimization is used to search for the optimal image features. The PCA and PSO-derived feature sets are then integrated to develop a hybrid set of features that are then used to train a Bayesian-based optimized support vector machine (SVM) and subspace discriminant ensemble-learning (SDEL) classifiers. The obtained results show improved classification performance for the ML classifiers trained by the proposed hybrid feature set over the original PCA, PSO, and all extracted feature sets for ALL multi-class classification. The Bayesian-optimized SVM trained with the proposed hybrid PCA-PSO feature set achieves the highest classification accuracy of 97.4%. The classification performance of the proposed feature engineering approach competes with the state of the art.

Comparison of ALEXNET and GOOGLENET convolutional neural network models to detect obstructive sleep apnea using single-channel electrocardiogram

Obstructive Sleep apnea (OSA) is a type of sleep disorder caused due to respiratory collapse during sleep. This sleep disorder generally goes undiagnosed and neglected. Severe OSA may cause arrhythmia, sudden death, high blood pressure, and other cardiac anomalies. Polysomnography (PSG) is the most popular gold standard used by many researchers to detect OSA. PSG required a well-equipped sleep laboratory and skilled persons to record multi- channel signals to detect OSA. PSG is a complex and expensive method and hence motivated to conduct the research using single-channel electrocardiogram (ECG). An automatic detection method of OSA using single-channel ECG in Convolutional Neural Network (CNN) takes less computing time as feature engineering does not require. This paper focuses on the automatic detection of OSA using ECG with two different deep CNN architectures AlexNet and GoogLeNet transfer learning. The apnea ECG datasheet is used for assessing the method proposed. The state of art using deep learning models are applied to single-channel ECG data. The GoogLeNet architecture is more complex and achieves 100% accuracy whereas AlexNet architecture shows 99.7% accuracy to detect OSA. The proposed work is applied to physionet apnea ECG online data which leads to an overfitting problem that can be resolved using clinical data to further enhance the robustness of the model.

A Deep Transfer Learning Framework for Sleep Stage Classification with Single-Channel EEG Signals.

The polysomnogram (PSG) is the gold standard for evaluating sleep quality and disorders. Attempts to automate this process have been hampered by the complexity of the PSG signals and heterogeneity among subjects and recording hardwares. Most of the existing methods for automatic sleep stage scoring rely on hand-engineered features that require prior knowledge of sleep analysis. This paper presents an end-to-end deep transfer learning framework for automatic feature extraction and sleep stage scoring based on a single-channel EEG. The proposed framework was evaluated over the three primary signals recommended by the American Academy of Sleep Medicine (C4-M1, F4-M1, O2-M1) from two data sets that have different properties and are recorded with different hardware. Different Time-Frequency (TF) imaging approaches were evaluated to generate TF representations for the 30 s EEG sleep epochs, eliminating the need for complex EEG signal pre-processing or manual feature extraction. Several training and detection scenarios were investigated using transfer learning of convolutional neural networks (CNN) and combined with recurrent neural networks. Generating TF images from continuous wavelet transform along with a deep transfer architecture composed of a pre-trained GoogLeNet CNN followed by a bidirectional long short-term memory (BiLSTM) network showed the best scoring performance among all tested scenarios. Using 20-fold cross-validation applied on the C4-M1 channel, the proposed framework achieved an average per-class accuracy of 91.2%, sensitivity of 77%, specificity of 94.1%, and precision of 75.9%. Our results demonstrate that without changing the model architecture and the training algorithm, our model could be applied to different single-channel EEGs from different data sets. Most importantly, the proposed system receives a single EEG epoch as an input at a time and produces a single corresponding output label, making it suitable for real time monitoring outside sleep labs as well as to help sleep lab specialists arrive at a more accurate diagnoses.

Development of Deep Neural Network Model for the Prediction of Road Crashes in Real Time

Road safety remains a global concern with the number of deaths and injury recorded from road traffic accidents estimated to be 1.5 million and 50 million respectively by 2025. Despite being predictable and largely preventable, the trend of road traffic crash is on the rise in Nigeria with an annual average of 33.7 deaths per 100,000 people. Proactive technique such as real time traffic and crash prediction has the potential to reduce the likelihood of crashes and to improve post-crash response. GoogleNet Convolutional Neural Network was developed in this study to classify road conditions and predict crashes along Ondo – Akure single carriage highway in Nigeria. Traffic flow relationships were established for the empirical data collected through video technique and compared to Green shields, Greenberg and Underwood models. The results were found generally satisfactory at an average coefficient of correlation of 0.96. The developed GoogleNet Convolutional network performed quite satisfactorily at predicting the probability of different traffic conditions – congested traffic (0.98), free-flowing traffic (0.64) and traffic crash (0.94). The developed algorithm can be integrated with traffic cameras and crowd-sourced images in areas that are not within the reach of surveillance cameras and sensors to report traffic condition in real time.

Deep convolutional neural networks for automated scoring of pentagon copying test results

This study aims to investigate the accuracy of a fine-tuned deep convolutional neural network (CNN) for evaluating responses to the pentagon copying test (PCT). To develop a CNN that could classify PCT images, we fine-tuned and compared the pre-trained CNNs (GoogLeNet, VGG-16, ResNet-50, Inception-v3). To collate our training dataset, we collected 1006 correct PCT images and 758 incorrect PCT images drawn on a test sheet by dementia suspected patients at the Osaka City Kosaiin Hospital between April 2009 and December 2012. For a validation dataset, we collected PCT images from consecutive patients treated at the facility in April 2020. We examined the ability of the CNN to detect correct PCT images using a validation dataset. For a validation dataset, we collected PCT images (correct, 41; incorrect, 16) from 57 patients. In the validation testing for an ability to detect correct PCT images, the fine-tuned GoogLeNet CNN achieved an area under the receiver operating characteristic curve of 0.931 (95% confidence interval 0.853–1.000). These findings indicate that our fine-tuned CNN is a useful method for automatically evaluating PCT images. The use of CNN-based automatic scoring of PCT can potentially reduce the burden on assessors in screening for dementia.

Research and Implementation of Flow Pattern Recognition for Gas-liquid Two-phase Flows Based on GoogLeNet

In order to solve the flow pattern recognition problem of gas-liquid two-phase flow in pipelines, this paper uses high-speed photography to sample the flow patterns of transparent pipe sections and combines the GoogLeNet convolutional neural network model under migration learning to implement a flow pattern recognition method with small samples. In this paper, the GoogLeNet Inception V1 network is used, and the convolutional layer and the pooling layer weights parameters obtained from its training on the imageNet dataset are retained, and the flow pattern samples obtained on the gas-liquid two-phase flow experimental platform are used to train the network model. The recognition accuracy was 98.37% with a training set of 400 and a test set of 100 samples of each flow type. The convolutional neural network directly uses images as data input without operations such as image pre-processing and feature extraction, and its unique fine-grained feature extraction enables the recognition of images by convolutional neural networks at a nearly human level.

Deep Learning Based Garbage Detection for Autonomous Garbage Collection Vehicles

With the rapid technological advances and the increasing human population, the need for more production has emerged and consumption has increased accordingly. As a result of this increased consumption, more garbage has been generated. Environmental pollution caused by these garbages emerges as a problem that people have to overcome both in Turkey and in the world. Many studies have been conducted to overcome this problem. Especially today, with the development of autonomous vehicles and artificial intelligence, the solutions using these technologies have increased. In this study, a new data set was created for autonomous garbage collection vehicles and a model was proposed in which these vehicles can be used. The data set was prepared with images of garbage with paper cups, which is one of the most polluting garbage, taken in different places, and images consisting of different garbage without paper cups. Paper cups were detected from these images with pre-trained Squenzenet, VGG-19 and GoogLeNet convolutional neural networks. The performance rate of the SquenzeNet, GoogLeNet and Vgg-19 networks used in the study was found as 97.77%, 96.44% 94.66%, respectively.

Improving time series features identification by means of Convolutional Neural Networks and Recurrence Plot

Within business forecasting, the choice of the forecasting model is crucial. In fact, not all models can be functional to all time series and their applicability is based on the features of the considered time series, such as trend, seasonality and randomness. The identification of features is, then, a fundamental activity to achieve the desired forecasting performance. On the other hand, time series features identification is known to be a time-consuming activity that requires the intervention of the decision maker. Hence, performing such a step of the forecasting process in an automated and reliable way would considerably improve the whole process. To this aim, this work proposes a method that combines the Recurrence Plot (RP) and the Convolutional Neural Networks (CNNs) to perform typical business time series features identification. RP is a visual tool that represents recurrence of the state space system by means of oriented structures in 2D images. CNNs are popular deep learning models that have achieved high recognition rates performing image classification tasks. Machine learning algorithms have been already successfully applied to the business forecasting process, limited to forecasting model selection step, while the CNN have been successfully applied in different fields, such as medicine, distinguishing ischemic disease in electroencephalogram images, not yet to business time series features identification. This work applies the GoogLeNet CNN to the RPs that represent simulated time series and measures the obtained classification accuracy by means of confusion matrix.

Big Data Analysis and Management of Healthcare Systems for Hacker Detection Based on Google Net Convolutional Neural Network

In recent times, Hacking has turn out to be more unfavorable than ever in all life fields, including the healthcare systems, with an increasing usage of information technology. By the expansion of technology development, the attacks number is too rising every few months in an exponential manner, which in turn makes the conventional IDS incapable to perceive. A healthcare system network intrusion detection method is proposed depending on the Google NET convolution neural network (Google NET). In healthcare system databases, intrusion detection (KDDs) can be seen as a search issue, which might be solved with the use of Google NET CNN algorithms. After pre-processing and characterizing the healthcare system data (including Electronic Health Records (EHR), Medical imaging data, Electronic Medical Records (EMR), etc.), the Google NET CNN model is used to simulate the intrusion into the healthcare system data. The low-level data intrusion is signified conceptually as the superior features with Google NET CNN, which in turn extracts the sample features separately, and by using MFO, network parameter is optimized (algorithm of optimization to meet the representation. At last, a sample test is conducted for the detection of healthcare system network intrusion behavior. The simulation outcome illustrate that the proposed technique has high accuracy on detection and a lower false-positive rate along with true positive rate.

A New Classification Method in Ultrasound Images of Benign and Malignant Thyroid Nodules Based on Transfer Learning and Deep Convolutional Neural Network

Thyroid nodule is a clinical disorder with a high incidence rate, with large number of cases being detected every year globally. Early analysis of a benign or malignant thyroid nodule using ultrasound imaging is of great importance in the diagnosis of thyroid cancer. Although the b-mode ultrasound can be used to find the presence of a nodule in the thyroid, there is no existing method for an accurate and automatic diagnosis of the ultrasound image. In this pursuit, the present study envisaged the development of an ultrasound diagnosis method for the accurate and efficient identification of thyroid nodules, based on transfer learning and deep convolutional neural network. Initially, the Total Variation- (TV-) based self-adaptive image restoration method was adopted to preprocess the thyroid ultrasound image and remove the boarder and marks. With data augmentation as a training set, transfer learning with the trained GoogLeNet convolutional neural network was performed to extract image features. Finally, joint training and secondary transfer learning were performed to improve the classification accuracy, based on the thyroid images from open source data sets and the thyroid images collected from local hospitals. The GoogLeNet model was established for the experiments on thyroid ultrasound image data sets. Compared with the network established with LeNet5, VGG16, GoogLeNet, and GoogLeNet (Improved), the results showed that using GoogLeNet (Improved) model enhanced the accuracy for the nodule classification. The joint training of different data sets and the secondary transfer learning further improved its accuracy. The results of experiments on the medical image data sets of various types of diseased and normal thyroids showed that the accuracy rate of classification and diagnosis of this method was 96.04%, with a significant clinical application value.

Power Efficient Design of High-Performance Convolutional Neural Networks Hardware Accelerator on FPGA: A Case Study With GoogLeNet

Convolutional neural networks (CNNs) have dominated image recognition and object detection models in the last few years. They can achieve the highest accuracies with several applications such as automotive and biomedical applications. CNNs are usually implemented by using Graphical Processing Units (GPUs) or generic processors. Although the GPUs are capable of performing the complex computations needed by the CNNs, their power consumption is huge compared to generic processors. Moreover, current generic processors are unable to cope up with the growing CNNs demand for computation performance. Therefore, hardware accelerators are the best choice to provide the required computation performance needed by the CNNs as well as affordable power consumption. Several techniques are adopted in hardware accelerators such as pruning and quantization. In this paper, a low-power dedicated CNN hardware accelerator is proposed based on GoogLeNet CNN as a case study. Weights pruning and quantization are applied to reduce the memory size by $57.6\times $ . Consequently, only FPGA on-chip memory is used for weights and activations storage without using offline DRAMs (Dynamic Random Access Memories). In addition, the proposed hardware accelerator utilizes zero DSP (Digital Signal Processing) units as all multiplications are replaced by shifting operations. The accelerator is developed based on a time-sharing/pipelined architecture, which processes the CNN model layer by layer. The architecture proposes a new data fetching mechanism that increases data reuse. Moreover, the proposed accelerator units are implemented in native RTL (Register Transfer Logic). The accelerator classifies 25.1 frames per second (fps) with 3.92W only, which is more power-efficient than other GoogLeNet implementations on FPGA in the literature. In addition, the proposed accelerator achieves an average classification efficiency of 91%, which is significantly higher than comparable architectures. Furthermore, this accelerator surpasses the popular CPUs such as Intel Core-i7 and GPUs such as GTX 1080Ti in terms of the number of frames processed per Watt.

GoogleNet CNN Neural Network towards Chest CT-Coronavirus Medical Image Classification

In the end of the year 2019 and the beginning of the year 2020, the world was overwhelmed by a medical pandemic that was not previously seen which is known Covid-19 (Coronavirus). Coronavirus (CoV) is a large family of viruses that cause illness ranging from the common cold to more severe diseases such as Middle East Respiratory Syndrome (MERS-CoV) and Severe Acute Respiratory Syndrome (SARS-CoV). This paper aims to improve the accuracy of detection for CT-Coronavirus images using deep learning for Convolutional Neural Networks (CNNs) that helps medical staffs for classification chest CT- Coronavirus medical image in early stage. Deep learning is successfully used as a tool for machine learning, where the CNNs are capable of automatically extracting and learning features medical image dataset. This research retrains GoogleNet CNN architecture over the COVIDCT-Dataset for classification CT- Coronavirus image. In this research, COVIDCT-Dataset contains 349 CT images containing clinical findings of COVID-19. The validation accuracy of retraining GoogleNet is 82.14% where elapsed time is 74 min and 37 sec.