VGG16 Convolutional Neural Network Research Articles

Target Detection and Network Optimization: Deep Learning in Face Expression Feature Recognition

Facial expression recognition technology has been more and more in demand in security, entertainment, education, medical, and other domains as artificial intelligence has advanced, and face expression recognition technology based on deep learning has become one of the research hotspots. However, there are still some issues with the existing deep learning convolutional neural network; the feature extraction technique has to be improved, and the design of the detailed network structure needs to be optimized. It is critical to do more research on the deep learning convolutional neural network model in order to increase the accuracy of face facial expression detection. In this paper, a deep learning convolutional neural network structure combining VGG16 convolutional neural network and long and short-term memory networks is designed to address the shortcomings of existing deep learning methods in face expression recognition, which are prone to overfitting and gradient disappearance, resulting in low test accuracy. This structure easily and effectively collects facial expression information and then classifies the retrieved features using a support vector machine to detect face expressions. Finally, the fer2013 dataset is used to train face expression recognition, and the results demonstrate that the built deep convolutional neural network model can effectively increase face expression identification accuracy.

Abstract 455: Automating integration of histomorphologic and immunohistochemistry data using computer vision tools

Abstract Background: Modern diagnostic pathology workflows involve the integration of histomorphologic, immunohistochemical (IHC), and molecular data to reach a final diagnosis. Recently, advances in deep learning have revolutionized pathology by providing the prospect for expert-level autonomous image analysis tools. Despite recent innovations in deep learning, integrating histomorphologic and molecular information found on respective H&E- and IHC-stained tissue sections still remains a challenge. Here, we aim to address this issue by incorporating computer vision tools, including deep learning and scale-invariant feature transform (SIFT), to align H&E-stained sections with accompanying IHC studies for automated subclassification of gliomas. Methods: To test the workflow, we trained the publicly available VGG19 convolutional neural network (CNN) using patches of pathologist-annotated H&E-stained WSIs to recognize histological patterns of 16 common tissue and brain tumor classes, including diffuse glioma, meningioma, metastatic carcinoma, and schwannoma. To complement the histomorphologic analysis, we optimized several deep learning classifiers, including Mask R-CNN, to recognize various IHC markers, such as Ki-67, IDH1-R132H and ATRX, relevant for molecular subclassification of gliomas. For the integrated analysis, we employed SIFT to align lesional regions of H&E and IHC images. Results: The histomorphologic classifier excelled at classification with accuracies of 100% for glioma, meningioma and metastatic carcinoma, and 93% for schwannoma (n = 125). The Mask R-CNN was tested on 147 images generated from 34 brain tumor Ki-67 WSIs and showed a high concordance with aggregate pathologists’ estimates (n = 3 assessors; y = 0.9712x -1.945, r = 0.9750). Using the deep learning classifiers and SIFT, we were able to identify tumor regions from the H&E images and align them to their corresponding IHC images. This resulted in significant improvement of ATRX and IDH1-R132H quantification compared to unaligned WSIs. Finally, SIFT and partially sampling the lesional regions decreased computational time significantly without compromising accuracy. Conclusion: SIFT can work in concert with deep learning tools to help examine the histomorphologic and molecular patterns of various brain tumors and provide a framework for integrating deep learning tools to provide automated diagnoses. Citation Format: Michael K. Lee, Kevin Faust, Madhumitha Rabindranath, Phedias Diamandis. Automating integration of histomorphologic and immunohistochemistry data using computer vision tools [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 455.

Optimal Slip Ratio Tracking Integral Sliding Mode Control for an EMB System Based on Convolutional Neural Network Online Road Surface Identification

As the main branch of the brake-by-wire system, the electro-mechanical brake (EMB) system is the future direction of vehicle brake systems. In order to enhance the vehicle braking effect and improve driver safety, a convolutional neural network (CNN) online road surface identification algorithm and an optimal slip ratio tracking integral sliding mode controller (ISMC) combined EMB braking control strategy is proposed in this paper. Firstly, according to the quarter-vehicle model and Burckhardt tire model, the vehicle braking control theory based on the optimal slip ratio is analyzed. Secondly, using the VGG-16 CNN method, an online road surface identification algorithm is proposed. Through a comparative study under the same dataset conditions, it is verified that the VGG-16 method has a higher identification accuracy rate than the SVM method. In order to further improve the generalization ability of VGG-16 CNN image identification, data enhancement is performed on the road surface image data training set, including image flipping, clipping, and adjusting sensitivity. Then, combined with the EMB system model, an exponential approach law method-based ISMC is designed to achieve the optimal slip ratio tracking control of the vehicle braking process. Finally, MATLAB/Simulink software is used to verify the correctness and effectiveness of the proposed strategy and shows that the strategy of real-time identifying road surface conditions through vision can make the optimal slip ratio of vehicle braking control reasonably adjusted, so as to ensure that the adhesion coefficient of wheel braking always reaches the peak value, and finally achieves the effect of rapid braking.

Label-free and automated approach to rapidly classify microsatellite instability (MSI) in early colon cancer (CC) analyzing the AIO ColoPredictPlus 2.0 (CPP) registry trial.

3616 Background: MSI due to mismatch repair defects accounts for 15-20% of all CC, has high prognostic and predictive value and is broadly utilized in treatment decisions. Artificial intelligence (AI) integrated, label-free quantum cascade laser (QCL) based infrared (IR) imaging resolves spatial and molecular alterations such as MSI in unstained cancer tissue sections. We aimed to evaluate the method for microsatellite instability/stability (MSI/MSS) classification in samples from the prospective multicenter AIO CPP registry trial. Methods: Paraffin-embedded unstained cancer tissue slides from patients (pts.) participating in CPP were measured (avg. 30 min/slide) and analyzed. The cohort was split into training (train), test (test), and validation (vali) sets. Cancer regions were first preselected based on a self-developed convolutional neural network (CNN) CompSegNet (Schuhmacher, medrxiv 2021). A VGG-16 CNN then classified MSI/MSS in these regions. Endpoints were area under receiver operating characteristic (AUROC) and area under precision recall curve (AUPRC). Results: 547 pts. (train n=331, test n=69, vali n=147) were analyzed. The baseline characteristics for the sub-cohorts are illustrated in the table. Mutation (MT) status: RAS MT: train 30% / test 30% / vali 37%; BRAF MT: train 27% / test 23% / vali 14%. The preselection of cancer regions reached a validation AUROC of 1.0. The subsequent MSI/MSS classifier reached a validation AUROC of 0.9 and AUPRC of 0.74 (sensitivity 85%, specificity 84%). Conclusions: Our multicenter approach using AI integrated label-free IR imaging provides an automated, fast, and reliable classification for MSI/MSS with an AUROC of 0.9 (sensitivity 85%, specificity 84%) almost comparable to the present gold standard immunohistochemistry. The method described here requires less samples for training when compared to other AI approaches which could facilitate the development of prognostic/predictive classifiers in the setting of randomized controlled trials. This novel technique may support further understanding of the increasingly important MSI CC cohort and support treatment decisions e.g. in specific subgroups such as targetable fusions. We expect our approach to be a broadly applicable diagnostic tool in the future.[Table: see text]

PD-11 In depth analysis of label-free infrared (IR) imaging-based microsatellite instability (MSI) classification in early colon cancer (CC) on samples from the AIO ColoPredictPlus 2.0 (CPP) registry trial

MSI occurs in 15-20% of all early CC due to a deficient mismatch repair system and testing is now considered standard for all colon cancers as MSI is of prognostic and predictive value. Presently, protein analysis using immunohistochemistry targeting the mismatch repair proteins is most frequently used in routine clinical pathology with high sensitivity. Alternatively, fragment length analysis of microsatellite loci or next-generation sequencing can be used. All the techniques referred to are time and tissue consuming. We utilize label-free quantum cascade laser (QCL) based IR imaging combined with artificial intelligence (AI) approaches to test unstained tissue samples for MSI. IR imaging is an emerging microscopic technique based on the interaction of electromagnetic waves with the molecules within the tissue creating molecular fingerprints. To establish the methodological robustness of our approach, we analyzed samples from the multicentric prospective AIO ColoPredict Plus (CPP) 2.0 registry trial. IR imaging can analyze unstained paraffin-embedded tissue slides within an average of 30 min/slide. All tissue samples were obtained from the CPP registry trial. For development of the MSI/MSS classification model the cohort was split into a training, test, and validation set. The training set was utilized for the optimization of a modified VGG-16 convolutional neural network (CNN) with area under receiver operating characteristic (AUROC) and area under precision recall curve (AUPRC) evaluated on the test set as endpoints. The final classification model is validated on the validation set and in depth analyzed for sub-cohorts (BRAF, KRAS, UICC stage, grading). A cohort of 547 patients (training n=331 (43% MSI), test n=69 (43% MSI), validation n=147 (18% MSI)) was selected from CPP. Baseline characteristics including BRAF mutations, were balanced between training and test set, thus providing a balanced ground-truth for the training of the classifier. By contrast, the validation set (18% MSI) corresponds to the natural occurrence of MSI, showing a typically pattern for MSI and MSS with a higher proportion of female patients, more right-sided cases and more BRAF mutations for MSI-samples. The selected MSI/MSS classifier reached a validation AUROC of 0.90 (sensitivity 85%, specificity 84%) and AUPRC of 0.74. The AUROC of the sub-cohorts range between 0.78 and 0.99 (BRAF: 0.93 (wt), 0.78 (mt); KRAS: 0.84 (wt), 0.99 (mt); Grading: 0.93 (G2), 0.84 (G3); UICC Staging: 0.93 (II), 0.86 (III)). AI integrated IR imaging demonstrates a reliable classification performance for MSI/MSS with an AUROC of 0.90 (sensitivity 85%, specificity 84%) and robustness over sub-cohorts e.g., BRAF, KRAS, or staging. It further approaches the performance of the present gold standard immunohistochemistry. The combination of spatial and biochemical information encoded in the IR images allows CNNs to track changes on the molecular level. Additionally, this method requires fewer samples than other AI approaches and maintains intact tissue for further molecular analysis. Overall, this method has the potential to become an applicable diagnostic tool beyond the scope of known biomarkers for tumor-agnostic assessments.

A 3D Fluorescence Classification and Component Prediction Method Based on VGG Convolutional Neural Network and PARAFAC Analysis Method

Three-dimensional fluorescence is currently studied by methods such as parallel factor analysis (PARAFAC), fluorescence regional integration (FRI), and principal component analysis (PCA). There are also many studies combining convolutional neural networks at present, but there is no one method recognized as the most effective among the methods combining convolutional neural networks and 3D fluorescence analysis. Based on this, we took some samples from the actual environment for measuring 3D fluorescence data and obtained a batch of public datasets from the internet species. Firstly, we preprocessed the data (including two steps of PARAFAC analysis and CNN dataset generation), and then we proposed a 3D fluorescence classification method and a components fitting method based on VGG16 and VGG11 convolutional neural networks. The VGG16 network is used for the classification of 3D fluorescence data with a training accuracy of 99.6% (as same as the PCA + SVM method (99.6%)). Among the component maps fitting networks, we comprehensively compared the improved LeNet network, the improved AlexNet network, and the improved VGG11 network, and finally selected the improved VGG11 network as the component maps fitting network. In the improved VGG11 network training, we used the MSE loss function and cosine similarity to judge the merit of the model, and the MSE loss of the network training reached 4.6 × 10−4 (characterizing the variability of the training results and the actual results), and we used the cosine similarity as the accuracy criterion, and the cosine similarity of the training results reached 0.99 (comparison of the training results and the actual results). The network performance is excellent. The experiments demonstrate that the convolutional neural network has a great application in 3D fluorescence analysis.

An FPGA Overlay for CNN Inference with Fine-grained Flexible Parallelism

Increasingly, pre-trained convolutional neural networks (CNNs) are being deployed for inference in various computer vision applications, both on the server-side in the data centers and at the edge. CNN inference is a very compute-intensive task. It is a challenge to meet performance metrics such as latency and throughput while optimizing power. Special-purpose ASICs and FPGAs are suitable candidates to meet these power and performance budgets simultaneously. Rapidly evolving CNN architectures involve novel convolution operations such as point convolutions, depth separable convolutions, and so on. This leads to substantial variation in the computational structure across CNNs and layers within a CNN. Because of this, FPGA reconfigurability provides an attractive tradeoff compared to ASICs. FPGA-based hardware designers address the structural variability issue by generating a network-specific accelerator for a single network or a class of networks. However, homogeneous accelerators are network agnostic and often sacrifice throughput and FPGA LUTs for flexibility. In this article, we propose an FPGA overlay for efficient processing of CNNs that can be scaled based on the available compute and memory resources of the FPGA. The overlay is configured on the fly through control words sent by the host on a per-layer basis. Unlike current overlays, our architecture exploits all forms of parallelism inside a convolution operation. A constraint system is employed at the host end to find out the per-layer configuration of the overlay that uses all forms of parallelism in the processing of the layer, resulting in the highest throughput for that layer. We studied the effectiveness of our overlay by using it to process AlexNet, VGG16, YOLO, MobileNet, and ResNet-50 CNNs targeting a Virtex7 and a bigger Ultrascale+VU9P FPGAs. The chosen CNNs have a mix of different types of convolution layers and filter sizes, presenting a good variation in model size and structure. Our accelerator reported a maximum throughput of 1,200 GOps/second on the Virtex7, an improvement of 1.2 \( \times \) to 5 \( \times \) over the recent designs. Also, the reported performance density, measured in giga operations per second per KLUT, is 1.3 \( \times \) to 4 \( \times \) improvement over existing works. Similar speed-up and performance density is also observed for the Ultrascale+VU9P FPGA.

Classification of Epileptic Brain Signals of College Students Based on Deep Learning

Epilepsy, as a common nervous system disease in the world, urgently needs effective methods for diagnosis and treatment. Scalp electroencephalogram (EEG) contains a lot of physiological and pathological information, which plays a very important role in the diagnosis of brain diseases such as epilepsy. At present, the clinical analysis of EEG signal is mainly based on the clinician’s visual analysis, which makes the clinician’s task heavy, and there is no quantitative standard for the analysis results. Therefore, the automatic classification of epileptic EEG signals has great potential in current clinical application. Based on the deep learning VGG16 convolutional neural network model, a method for automatic detection of seizure signals is proposed in this paper. Firstly, the samples are preprocessed, then enter the VGG16 convolutional neural network model for training, and finally input the validation set for validation. The experimental results indicate that this model can distinguish the pre-ictal EEG, interictal EEG and ictal EEG.

Retrieval of the planetary boundary layer height from lidar measurements by a deep-learning method based on the wavelet covariance transform.

Understanding and characterization of the planetary boundary layer (PBL) are of great importance in terms of air pollution management, weather forecasting, modelling of climate change, etc. Although many lidar-based approaches have been proposed for the retrieval of the PBL height (PBLH) in case studies, development of a robust lidar-based algorithm without human intervention is still of great challenging. In this work, we have demonstrated a novel deep-learning method based on the wavelet covariance transform (WCT) for the PBLH evaluation from atmospheric lidar measurements. Lidar profiles are evaluated according to the WCT with a series of dilation values from 200 m to 505 m to generate 2-dimensional wavelet images. A large number of wavelet images and the corresponding PBLH-labelled images are created as the training set for a convolutional neural network (CNN), which is implemented based on a modified VGG16 (VGG - Visual Geometry Group) convolutional neural network. Wavelet images obtained from lidar profiles have also been prepared as the test set to investigate the performance of the CNN. The PBLH is finally retrieved by evaluating the predicted PBLH-labelled image and the wavelet coefficients. Comparison studies with radiosonde data and the Micro-Pulse-Lidar Network (MPLNET) PBLH product have successfully validated the promising performance of the deep-learning method for the PBLH retrieval in practical atmospheric sensing.

Emotion classification of human facial images by a neural network

Facial recognition using artificial neural networks is a biometric technology currently being used in fields such as cybersecurity and criminal investigation. We sought to automatically distinguish between an image of a happy human face and an image of a sad human face with predictions that are better than random guesses. We trained a machine learning model (VGG16, a type of convolutional neural network) on a public image dataset of 12,000 human faces. The resulting model predicted the emotion label 93% of time when shown a test set of 200 images that the model had never seen during training. The results show that the VGG16 convolutional neural network learned features from our data that produced an output which was sufficient to train the additional layers of the model to perform our task at 93% accuracy. We suspect this is because VGG16 was already familiar with features that it learned from ImageNet (a separate dataset of over 2 million images). It is currently unknown whether our model can predict labels for new images outside of the FER13 dataset, but preliminary tests show promising results. Our model was trained using a cloud computing service and a relatively small amount of data, indicating that these kinds of results are easily obtainable by all.

Fast and automated biomarker detection in breath samples with machine learning

Volatile organic compounds (VOCs) in human breath can reveal a large spectrum of health conditions and can be used for fast, accurate and non-invasive diagnostics. Gas chromatography-mass spectrometry (GC-MS) is used to measure VOCs, but its application is limited by expert-driven data analysis that is time-consuming, subjective and may introduce errors. We propose a machine learning-based system to perform GC-MS data analysis that exploits deep learning pattern recognition ability to learn and automatically detect VOCs directly from raw data, thus bypassing expert-led processing. We evaluate this new approach on clinical samples and with four types of convolutional neural networks (CNNs): VGG16, VGG-like, densely connected and residual CNNs. The proposed machine learning methods showed to outperform the expert-led analysis by detecting a significantly higher number of VOCs in just a fraction of time while maintaining high specificity. These results suggest that the proposed novel approach can help the large-scale deployment of breath-based diagnosis by reducing time and cost, and increasing accuracy and consistency.

Woody Plant Encroachment: Evaluating Methodologies for Semiarid Woody Species Classification from Drone Images

Globally, native semiarid grasslands and savannas have experienced a densification of woody plant species—leading to a multitude of environmental, economic, and cultural changes. These encroached areas are unique in that the diversity of tree species is small, but at the same time the individual species possess diverse phenological responses. The overall goal of this study was to evaluate the ability of very high resolution drone imagery to accurately map species of woody plants encroaching on semiarid grasslands. For a site in the Edwards Plateau ecoregion of central Texas, we used affordable, very high resolution drone imagery to which we applied maximum likelihood (ML), support vector machine (SVM), random forest (RF), and VGG-19 convolutional neural network (CNN) algorithms in combination with pixel-based (with and without post-processing) and object-based (small and large) classification methods. Based on test sample data (n = 1000) the VGG-19 CNN model achieved the highest overall accuracy (96.9%). SVM came in second with an average classification accuracy of 91.2% across all methods, followed by RF (89.7%) and ML (86.8%). Overall, our findings show that RGB drone sensors are indeed capable of providing highly accurate classifications of woody plant species in semiarid landscapes—comparable to and even greater in some regards to those achieved by aerial and drone imagery using hyperspectral sensors in more diverse landscapes.

Machine Vision and Intelligent Algorithm Based on Neural Network.

Neural network algorithms and intelligent algorithms are hot topics in the field of deep learning. In this study, the neural network algorithm and intelligence are optimized, and it is used in simulation experiments to improve the target image recognition ability of the algorithm in the machine vision environment. First, this paper introduces the application of neural networks in the field of machine vision. Second, in the experiment, the improved VGG-16 convolutional neural network (CNN) model is applied to metal block defect detection. Experimental results show that the optimized network can classify metal block defects with the maximum accuracy of 99.28%. Then, the intelligent algorithm based on neural network is studied, and the CIFAR-10 data set is taken as the experimental target for training test and verification test. Using BP algorithm, particle swarm optimization algorithm (PSO-BP), and improved neural network algorithm, respectively, the convergence speed of ICS algorithm based on BP neural network is compared. In contrast, ICS-BP algorithm has the fastest convergence speed and converges when the number of iterations is 32, followed by PSO-BP algorithm.

Use of deep learning in the MRI diagnosis of Chiari malformation type I

PurposeTo train deep learning convolutional neural network (CNN) models for classification of clinically significant Chiari malformation type I (CM1) on MRI to assist clinicians in diagnosis and decision making.MethodsA retrospective MRI dataset of patients diagnosed with CM1 and healthy individuals with normal brain MRIs from the period January 2010 to May 2020 was used to train ResNet50 and VGG19 CNN models to automatically classify images as CM1 or normal. A total of 101 patients diagnosed with CM1 requiring surgery and 111 patients with normal brain MRIs were included (median age 30 with an interquartile range of 23–43; 81 women with CM1). Isotropic volume transformation, image cropping, skull stripping, and data augmentation were employed to optimize model accuracy. K-fold cross validation was used to calculate sensitivity, specificity, and the area under receiver operating characteristic curve (AUC) for model evaluation.ResultsThe VGG19 model with data augmentation achieved a sensitivity of 97.1% and a specificity of 97.4% with an AUC of 0.99. The ResNet50 model achieved a sensitivity of 94.0% and a specificity of 94.4% with an AUC of 0.98.ConclusionsVGG19 and ResNet50 CNN models can be trained to automatically detect clinically significant CM1 on MRI with a high sensitivity and specificity. These models have the potential to be developed into clinical support tools in diagnosing CM1.

Forensic bone age estimation of adolescent pelvis X-rays based on two-stage convolutional neural network.

In the forensic estimation of bone age, the pelvis is important for identifying the bone age of teenagers. However, studies on this topic remain insufficient as a result of lower accuracy due to the overlapping of pelvic organs in X-ray images. Segmentation networks have been used to automate the location of key pelvic areas and minimize restrictions like doubling images of pelvic organs to increase the accuracy of estimation. This study conducted a retrospective analysis of 2164 pelvis X-ray images of Chinese Han teenagers ranging from 11 to 21years old. Key areas of the pelvis were detected with a U-Net segmentation network, and the findings were combined with the original X-ray image for regional augmentation. Bone age estimation was conducted with the enhanced and not enhanced pelvis X-ray images by separately using three convolutional neural networks (CNNs). The root mean square errors (RMSE) of the Inception-V3, Inception-ResNet-V2, and VGG19 convolutional neural networks were 0.93years, 1.12years, and 1.14years, respectively, and the mean absolute errors (MAE) of these networks were 0.67years, 0.77years, and 0.88years, respectively. For comparison, a network without segmentation was employed to conduct the estimation, and it was found that the RMSE of the three CNNs above became 1.22years, 1.25years, and 1.63years, respectively, and the MAE became 0.93years, 0.96years, and 1.23years. Bland-Altman plots and attention maps were also generated to provide a visual comparison. The proposed segmentation network can be used to reduce the influence of restrictions like image overlapping of organs and can thus increase the accuracy of pelvic bone age estimation.

A deep learning approach for real time process monitoring and curling defect detection in Selective Laser Sintering by infrared thermography and convolutional neural networks

Selective Laser Sintering (SLS) of polyamide powders is one of the most prevalent additive manufacturing (AM) processes. Failures induced by curling and other process irregularities affect mechanical properties and part quality. In order to improve cost- and resource-efficiency, flexibility and sustainability of SLS, a real time and in-situ quality control of the whole SLS process chain is needed. Machine learning (ML) and especially deep learning (DL) is increasingly used for SLS quality control in recent time. In this approach, we investigated applications of DL to implement an in-situ quality control of the SLS process. Especially convolutional neural networks (CNN) have been used to classify infrared thermography recordings containing artificially induced defects. Using VGG16 CNN with the thermal recordings as the input data, an average curling failure detecting accuracy of 98,54 % was achieved. These results encourage the deployment of DL for non-destructive, in-situ quality control of SLS processes.

Learning Aided System for Agriculture Monitoring Designed Using Image Processing and IoT-CNN

The Internet of Things (IoT) and artificial intelligence (AI) based methods for monitoring, control, and decision support are combined to design of a smart agriculture assistance system. The proposed system has a sensor pack that provides continuous data capture of temperature records, air and soil moisture and a camera for obtaining near-infrared (NIR) images of the plant leaves for use with an AI decision support system. We identify twelve types of vegetation for the study, out of which five disease classes of the tomato leaves are categorized using a Convolutional Neural Network (CNN). The work also includes experiments conducted with multiple clustering-based segmentation methods and some features namely Gray level co-occurrence matrix (GLCM), Local binary pattern (LBP), Local Binary Gray Level Co-occurrence Matrix (LBGLCM), Gray Level Run Length Matrix (GLRLM), and Segmentation-based Fractal Texture Analysis (SFTA). Out of several AI tools, CNN proves to be effective in providing automated decision support for classifying the plant leaf disease types through a cloud server that can be accessed using an app. Extensive on-field trials show that the system (VGG16 CNN, GLCM and a fuzzy based clustering) is effective in hot and humid conditions and proves to be reliable in identifying disease classes of certain vegetable types, certain usable vegetation cover of farmland and regulation of watering mechanism of crops.

Распознавание микроскопических изображений пыльцевых зерен с помощью сверточной нейронной сети VGG-16

The article presents the result of an experiment on the application of transfer learning using the Visual Geometry Group with 16 layers (VGG-16) convolutional neural network in relation to the problem of recognizing pollen grains in images. An analysis of the information-theoretical base on the application of machine learning algorithms to the problem of classifying pollen grains over the past few years has shown the need to develop (apply) a new method for recognizing images of pollen grains obtained using an optical microscope. Currently, automatic classification for pollen identification is becoming a very active area of research. The article substantiates the task of automating the classification of pollen grains. The aim of the study is to analyze the efficiency and accuracy of classifying microscopic images of pollen grains using transfer learning of the VGG-16 convolutional neural network. Transfer learning was performed using the VGG-16 neural network, which has 13 convolutional layers grouped into 5 blocks with pooling and 3 smoothing layers at the output. Since transfer learning is used, the number of training epochs can be chosen to be small. For this network, only the smoothing output layers change, and the feature extraction remains the same as in the classical model. Therefore, it was chosen to use 10 training epochs. Other hyperparameters are as follows: Drop Out regularization with a probability of 50%, optimization method is ADAM, activation function is sigmoid, loss function is cross-entropy, batch size is 32 images. As a result, by adjusting the hyperparameters of the model and using augmentation, it was possible to achieve a share of correct recognitions of about 80%. At the same time, due to the different number of training examples, the particular characteristics of the classes differ somewhat. Thus, the maximum precision and recall reach 94% and 83%, respectively, for the Dandelion type. In the future, studies are planned to use augmentation as a preprocessing to create a balanced sample. By applying the VGG-16 convolutional neural network to the problem of pollen grain image recognition, high accuracy and efficiency of the method were achieved.

Evaluation of Parameter Fine-Tuning with Transfer Learning for Osteoporosis Classification in Knee Radiograph

Osteoporosis is a bone disease that raises the risk of fracture due to the density of the bone mineral being low and the decline of the structure of bone tissue. Among other techniques, such as Dual-Energy X-ray Absorptiometry (DXA), 2D x-ray pictures of the bone can be used to detect osteoporosis. This study aims to evaluate deep convolutional neural networks (CNNs), applied with transfer learning techniques, to categorize specific osteoporosis features in knee radiographs. For objective labeling, we obtained a selection of patient knee x-ray images. The study makes use of the Visual Geometry Group Deep (VGG-16), and VGG-16 with fine-tuning. In this work, the deployed CNNs were assessed using state-of-the-art metrics such as accuracy, sensitivity, and specificity. The evaluation shows that fine-tuning enhanced the VGG-16 CNN's effectiveness for detecting osteoporosis in radiographs of the knee. The accuracy of the VGG-16 with parameter fine-tuning was 88% overall, while the accuracy of the VGG-16 without parameter fine-tuning was 80%.

Smart Gravimetric System for Enhanced Security of Accesses to Public Places Embedding a MobileNet Neural Network Classifier

In this article, we present a smart gravimetric system for the automatic security monitoring of the accesses to public places with some entrance ticket or pass required (e.g., railway stations, subway stations, museums, exhibitions). The main objective is to spot illegal behaviors, for example, two people trying to enter using only one ticket; simultaneously, other events of interest can also be detected, for example, the passage of a person with disabilities or a person with a stroller. These pieces of information can be used to arrange the appropriate assistance or to conduct statistical analyses on the monitored place frequentation. The proposed system is based on a robust gravimetric footboard, composed of steel plates laid on load cells. The data processing, classification included, is totally executed on embedded STM32 microcontrollers, following the edge-computing paradigm. The classification exploits a MobileNetV2, a class of neural networks for computer vision, especially created for embedded and mobile applications. We realized a prototype of the gravimetric system for laboratory testing and we acquired a set of samples for the considered event classes to assess the MobileNet network capabilities. We compared the performance of this implementation with a more classic VGG16 convolutional neural network (CNN) running on a PC machine. We show that in the face of akin classification results and similar execution times between the two, the embedded MobileNet version brings many advantages, namely portability, cost, size, energy consumption reduction, and finally simplification in the installation procedure.