SPOT Images Research Articles

High Resolution of Plasmonic Resonance Scattering Imaging with Deep Learning.

The dark-field microscopy (DFM) imaging technology has the advantage of a high signal-to-noise ratio, and it is often used for real-time monitoring of plasmonic resonance scattering and biological imaging at the single-nanoparticle level. Due to the limitation of the optical diffraction limit, it is still a challenging task to accurately distinguish two or more nanoparticles whose distance is less than the diffraction limit. Here, we propose a computational strategy based on a deep learning framework (NanoNet), which will realize the effective segmentation of the scattered light spots in diffraction-limited DFM images and obtain high-resolution plasmonic light scattering imaging. A small data set of DFM and the corresponding scanning electron microscopy (SEM) image pairs are used to learn for obtaining a highly resolved semantic imaging model using NanoNet, and thus highly resolved DFM images matching the resolution of those acquired using SEM can be obtained. Our method has the ability to transform diffraction-limited DFM images to highly resolved ones without adding a complex optical system. As a proof of concept, a highly resolved DFM image of living cells through the NanoNet technique is successfully made, opening up a new avenue for high-resolution optical nanoscopic imaging.

Uncertainty-aware skin cancer detection: The element of doubt

Artificial intelligence (AI)-based medical diagnosis has received huge attention due to its potential to improve and accelerate the decision-making process at the patient level in a range of healthcare settings. Despite the recent signs of progress in this field, reliable quantification and proper communication of predictive uncertainties have been fully or partially overlooked in the existing literature on AI applications for medical diagnosis. This paper studies the automatic diagnosis of skin cancer using dermatologist spot images. Three different uncertainty-aware training algorithms (MC dropout, Bayesian Ensembling, and Spectral Normalized Neural Gaussian Process) are utilized to detect skin cancer. The performances of the three above-mentioned algorithms are compared from different perspectives. In addition, some images from the Cifar10 dataset are applied as the out-of-domain data and the performances of the algorithms are evaluated and compared for images that are far from the training samples. The accuracy, uncertainty accuracy, uncertainty accuracy for out-of-domain distribution samples, and the uncertainties of the predictions are reported in all cases and compared.

Ultrasound Image under Artificial Intelligence Algorithm in Thoracoscopic Surgery for Papillary Thyroid Carcinoma

This study aimed to investigate the therapeutic effect of thoracoscopic thyroidectomy guided by an ultrasound image based on an artificial intelligence algorithm on papillary thyroid carcinoma. Patients diagnosed with papillary thyroid carcinoma by imaging examination or needle biopsy in hospital were selected. The subjects were randomly divided into the experimental group and the control group. In the experimental group, 94 patients underwent conventional thoracoscopic thyroid tumor resection. In the control group, 119 patients underwent thoracoscopic thyroidectomy under ultrasonic guidance based on an intelligent algorithm. The adoption effect of ultrasound imaging based on the least variance algorithm in thoracoscopic papillary thyroid carcinoma surgery was evaluated by comparing the differences between the two groups of patients after laparoscopic thyroidectomy. The results showed that the ultrasonic imaging resolution and sound-absorbing spot imaging effect based on an artificial intelligence algorithm were superior to those of conventional ultrasonic imaging. Compared with the control group, the average duration of surgery in the experimental group was reduced and the intraoperative blood loss, postoperative cumulative drainage volume, postoperative drainage tube duration, hospital stay, and pain evaluation were dramatically better ( P < 0.05 ). It was confirmed that ultrasound-guided endoscopic thyroid surgery based on the least variance algorithm had better results than conventional surgery and showed the same safety as traditional surgery, which was of clinical promotion value. However, some evaluations were influenced by subjective judgment, and the accuracy of data conclusions needed to be further studied.

Intelligent System for Detecting Tooth Decay in OPG Images by Image Processing

Introduction: Dental caries and lesions are one of the most common dental diseases worldwide, affecting a high percentage of the world's population. Research has shown that more than 90% of adults have experienced tooth decay, so early detection of dental lesions in the early stages is a major aspect of maintaining dental health.Materials and Methods: In order to diagnose decayed teeth, first, the difference between decayed teeth and healthy teeth was discussed. By first calling the OPG images and removing the inappropriate margins of the image. Then the gray-scale image was converted to binary and after removing the noise, thresholding was performed. Then we found the location of the white spots in the black image and displayed the processed image on the user interface.Results: Computer image processing systems have been developed to be able to process and analyze images with better speed and accuracy. The proposed system has a good performance in diagnosing dental caries and the results show that the proposed system is reliable and optimal and as a clinical decision support system can help dentists in accurate and early diagnosis of dental caries at the initial level of caries.Conclusion: In this article, using the advancement of technology that has been formed in recent years, we were able to provide a solution to facilitate medical science. With the science of image processing, eye errors that may cause problems during the diagnosis and recording of the disease in the process of recovery are significantly reduced.

Land cover classification from very high spatial resolution images via multiscale object-driven CNNs and automatic annotation

Very high spatial resolution (VHR) images are accessible for land-cover classification. Because of the complex information and deep features associated with increased spatial resolution, it is difficult to find an effective method for improving the classification accuracy of finer-resolution remote sensing images. We propose a scheme for sample selection and sample autoannotation to establish a new labeled benchmark using vector maps. In addition, a suitable multiscale object-driven convolutional neural networks (multi-OCNNs) were developed for the new benchmark. We trained multi-OCNNs to capture the deep and contextual information contained in samples of the VM-RSI benchmark to predict the land cover categories from target region images with no labels. Furthermore, an object-driven voting strategy applied in multi-OCNNs provides clear boundary information of ground objects and less noise in CNN predictions. Experiments on multisource VHR images, including SPOT-6, Gaofen-2, and ALOS satellite images, show that the proposed approach outperforms other object-based algorithms by achieving good performance results, verifying its feasibility, and effectiveness for land cover classification.

Optimal design of photon-sieve holograms for fine-line image generation

A photon-sieve hologram (PSH) is a computer-generated hologram that can generate fine-line images. As a high-resolution image can be projected directly without using a reducing optical system, it can also be used for short-wavelength radiation, such as x-rays. We propose an optimum design method for the PSH so that a high-quality line image can be projected. We optimised straight-line and L-shaped bent-line images using numerical simulation based on vector diffraction theory and conjugate-gradient methods. This optimisation helped homogenise the intensity of the image and remove unwanted side lobes and spots around the image. We also experimentally examined the bent-line image before and after the optimisation to confirm its effect.

Imaging and reconstruction of high-energy proton beam spot before target based on secondary gamma rays.

The distribution and stability of the incident proton beam spot are critical for the stable operation of a high-power spallation target. This study proposes a method to capture images of the incident proton beam spot based on secondary gamma rays. The distribution of the backward secondary gamma rays irradiated by the primary proton beam on the incident surface of the target is close to the distribution of the protons and can be measured at a low-radiation position far from the target area. A relation of distributions between the incident protons and the secondary gammas is constructed by using the point response function of this pinhole imaging system. The proposed method of imaging is suitable for monitoring the distribution of the proton beam on the target in facilities that use a beam power of several megawatts or tens of megawatts, such as spallation neutron sources or accelerator-driven subcritical systems.

Comparison of choroidal hyperreflective spots on optical coherence tomography images between both eyes of normal subjects.

Advancement of optical coherence tomography (OCT) technology allows for better in vivo visualization of the choroidal architecture, which comprises vessels and stroma. However, most OCT studies using image binarization methods have focused only on choroidal vessels represented by dark pixels. This study aimed to compare the distribution of choroidal hyperreflective spots on swept-source OCT (SS-OCT) images between both eyes of normal subjects. In this observational comparative study, we included SS-OCT images of healthy subjects, which were prospectively obtained to compare images among the devices. SS-OCT images acquired using PLEX Elite 9000 and DRI-OCT Triton were analyzed. En-face OCT images were obtained at five different depth positions of the inner choroid at the macula. The mean reflectivity of the choroidal slabs, the number, total area, and circularity of hyperreflective spots were quantitatively compared between the devices and between both eyes of the same subjects. In 30 eyes of 15 healthy subjects, the mean reflectivity of the choroidal slabs varied with the scan depth on both devices (P<0.001 and P<0.001). Hyperreflective spots were similarly distributed in the images from both devices, but at different depths. The number and area of hyperreflective spots in the second and third layers of the DRI-OCT Triton were positively correlated with those in the fourth and fifth layers of the PLEX Elite 9000, respectively (all P<0.05). The intraclass correlation coefficients (ICC) for the area of hyperreflective spots were excellent for the third slab of the DRI-OCT and the fifth slab of the PLEX Elite (ICC =0.798; 95% CI, -0.576-0.904). The number and area of hyperreflective spots were correlated between both eyes in the third, fourth and fifth layers on DRI-OCT Triton (all P<0.05) and in the first, second, and fifth layers on PLEX Elite 9000 (all P<0.05). Hyperreflective spots in en-face images from two different SS-OCT devices were similarly observed between both eyes of the same person. The distributions of spots between the two eyes of the same person were correlated. These findings suggest that the distribution of hyperreflective spots on the choroid reflects the choroidal characteristics of the subject.

Research on proton beam spot imaging based on pixelated gamma detector

The secondary particles from the spallation target of the China Spallation Neutron Source are mainly gammas and neutrons, which are related to the incident proton. The reconstruction of the proton beam spot could be implemented based on the distribution of the positions of secondary gammas or neutrons. The methods of pinhole imaging and Compton imaging are developed by measuring the position distribution of gammas based on the pixelated detector. The secondary gammas could be detected by the pixelated gamma detector directly. The neutron can be identified by detecting the characteristic (478 keV) γ-rays from the 10B(n, α) reactions. In order to detect secondary neutrons, a layer of 10B converter is added before the pixelated gamma detector. The pixelated gamma detector is sensitive to the characteristic (478 keV) γ-rays and then the neutron imaging could be achieved based on measuring the position distribution of the characteristic (478 keV) γ-rays.

A large dataset of white blood cells containing cell locations and types, along with segmented nuclei and cytoplasm

Accurate and early detection of anomalies in peripheral white blood cells plays a crucial role in the evaluation of well-being in individuals and the diagnosis and prognosis of hematologic diseases. For example, some blood disorders and immune system-related diseases are diagnosed by the differential count of white blood cells, which is one of the common laboratory tests. Data is one of the most important ingredients in the development and testing of many commercial and successful automatic or semi-automatic systems. To this end, this study introduces a free access dataset of normal peripheral white blood cells called Raabin-WBC containing about 40,000 images of white blood cells and color spots. For ensuring the validity of the data, a significant number of cells were labeled by two experts. Also, the ground truths of the nuclei and cytoplasm are extracted for 1145 selected cells. To provide the necessary diversity, various smears have been imaged, and two different cameras and two different microscopes were used. We did some preliminary deep learning experiments on Raabin-WBC to demonstrate how the generalization power of machine learning methods, especially deep neural networks, can be affected by the mentioned diversity. Raabin-WBC as a public data in the field of health can be used for the model development and testing in different machine learning tasks including classification, detection, segmentation, and localization.

F-18 DOPA PET/CT in pediatric patients with hyperinsulinemic hypoglycemia: A correlation with genetic analysis.

Persistent hyperinsulinemic hypoglycemia is a rare but life-threatening disease of infancy and childhood. 18F-DOPA PET/CT has been shown to be a useful modality in the localization of focal pancreatic lesions in these patients. We aimed to assess the role of 18F-DOPA PET/CT in such patients at our institution. In this prospective study, 18F-DOPA PET/CT scans and clinical details of 20 children with clinical diagnosis of hyperinsulinemic hypoglycemia were reviewed. Scans were acquired at 5 min postinjection of 2-3 mCi of 18F-DOPA on dedicated PET/CT scanners (Biograph mCT, Siemens Inc and Discovery PET/CT, GE). Abdominal spot images over 1-2 bed positions were acquired. Additionally, genetic mutation status, where available, was correlated to the scan findings. Out of 20 children (7 female and 13 male), 13 were infants. The age of the children ranged from 3 months to 8 years. Fifteen children had undergone gene analysis, 12 were positive for mutation in ABCC8, 1 for GLUD-1, 1 for GCK mutations and 1 had not shown any mutation. 18F-DOPA PET/CT scan showed 5 focal pancreatic lesions in 5 children (1 in each), two focal lesions in 1 child and diffuse pancreatic uptake in 14 children. 18F-DOPA PET/CT is a useful modality for localizing focal pancreatic lesions in children with persistent hyperinsulinemic hypoglycemia. The detection rate is significantly higher in patients with ABCC8 paternal monoallelic recessive gene mutation. 18F-DOPA PET/CT scan consequent to findings on genetic analysis appears to be useful in planning the management of children with hyperinsulinism.

An Automatic Procedure for Overheated Idler Detection in Belt Conveyors Using Fusion of Infrared and RGB Images Acquired during UGV Robot Inspection

Complex mechanical systems used in the mining industry for efficient raw materials extraction require proper maintenance. Especially in a deep underground mine, the regular inspection of machines operating in extremely harsh conditions is challenging, thus, monitoring systems and autonomous inspection robots are becoming more and more popular. In the paper, it is proposed to use a mobile unmanned ground vehicle (UGV) platform equipped with various data acquisition systems for supporting inspection procedures. Although maintenance staff with appropriate experience are able to identify problems almost immediately, due to mentioned harsh conditions such as temperature, humidity, poisonous gas risk, etc., their presence in dangerous areas is limited. Thus, it is recommended to use inspection robots collecting data and appropriate algorithms for their processing. In this paper, the authors propose red-green-blue (RGB) and infrared (IR) image fusion to detect overheated idlers. An original procedure for image processing is proposed, that exploits some characteristic features of conveyors to pre-process the RGB image to minimize non-informative components in the pictures collected by the robot. Then, the authors use this result for IR image processing to improve SNR and finally detect hot spots in IR image. The experiments have been performed on real conveyors operating in industrial conditions.

THE CONTRIBUTION OF RADAR REMOTE SENSING VIA SENTINEL-1 DATA AND PHOTO-INTERPRETATION BY GOOGLE EARTH IMAGES FOR WETLAND MAPPING

Abstract. Wetlands are considered as sensitive ecosystems exposed and threatened by climate change and the urbanization of natural environments. In the purpose of managing these sensitive areas and conservatizing their biodiversity, remote sensing is an efficient way to track environmental variables over large areas as wetlands. However, when it comes to the study of hydrologic dynamics, high temporal and spatial resolutions are essential. Since the access to optical satellite imagery is restrictive because of the large cloud cover that masks the ground, radar sensors that are working in the microwave field, are particularly suited to the characterization of hydrological dynamics due to the sensitivity of their measurements in the presence of water, regardless of the vegetation in place. Recently, radar remote sensing has experienced a real revolution with the launch of the Sentinel-1A satellite in 2014, followed by its twin Sentinel-1B two years later by the European Space Agency as part of the Copernicus program. These sensors acquire C-band data (λ = 5.6 cm) with a temporal resolution of 12 days by satellite and their distribution is open and free. This article aims to assess the potential of Sentinel A1 SAR data for wetland mapping in the city of Benslimane (Central Morocco). The first part is explaining the methodology for mapping water surfaces. We identified a confusion of the C-band radar response of water surfaces and that of certain bare soils. We then showed that the VH polarization is the most suitable for the mapping of water surfaces, comparing four methods of detecting areas in water. It. The second part is discussing the use of unsupervised methods without a priori data demonstrating that the methods taking into account the spatial neighborhood give better results. Temporal filtering has been developed and has made it possible to improve detection and to overcome confusion between bare soil and permanent water surfaces. Water surfaces larger than 0.5 ha are at 80% detected. Classification was performed using the SVM (Support Vector Machine) algorithm. This latter information was then implemented into the thematic map derived from SPOT-4 images to obtain the final weltands map.

Specific pattern of bone scan as a result of unintentional-intraarterial injection into the ulnar artery.

Background:Bone scintigraphy with 99mTc labeled radiopharmaceuticals is a valuable method in nuclear medicine for assessing the bony structure. In clinical setting, bone scintigraphy is accomplished after the injection of 99mTc labeled diphosphonate complexes into a peripheral vein. Inadvertent intra-arterial injection on the antecubital region may cause a special form of artifacts leading to problems inaccurately interpreting these studies as functional images.Case Presentation:We present a 44-year-old man with history of chest wall pain for bone scintigraphy as part of a work-up for determining the pain source. The patient received an injection of 740MBq 99mTc-methylene diphosphonate (MDP) into a blood vessel at the right forearm. Two hours later, an increased uptake of activity was observed on the right forearm and ulnar half of the wrist-hand in the whole body and spot images. The scan findings were consistent with the anatomical and physiological expectations of the ulnar arterial perfusion range. This case displays that an incidental injection with a 99mTc labeled diphosphonate into the ulnar artery results in a hot ulnar half of the palm and ulnar-sided three digits, because these areas are directly exposed to radiopharmaceutical, therefore more radiopharmaceutical is deposited.Conclusion:It is important for the physician, and especially for the nuclear medicine technician, to know the specific appearance created in the bone scan due to such an event so that they do not make a diagnostic mistake.

Improved Centroid Algorithm for Beam Stability Control System

In order to eliminate the beam jitter caused by the air disturbance and the vibrating devices, the error of light spotted jitter must be detected firstly. The beam jitter error can be corrected by a compensator to realize the beam stability control system. In this paper, an Improved Centroid Algorithm (ICA) combining with the Gaussian Fitting Algorithm (GFA), the Bilinear Interpolation and the Weighted Center of Mass (WCOM) was proposed. A Gaussian template is used to roughly locate the spot image in order to improve the efficiency and stability of the algorithm. The light spot image is interpolated by the Bilinear Interpolation to improve the accuracy of centroid identification. The identification accuracy and efficiency of the algorithm were verified by 100 sets of light spot experiments. The experimental results showed that the identification accuracy and efficiency of the Improved Centroid Algorithm were higher than those of the other algorithms. Our algorithm was of great significance for improving the performance of the beam stability control system.

Evaluation Model of Regional Comprehensive Disaster Reduction Capacity under Complex Environment.

In order to realize the evaluation of regional comprehensive disaster reduction capacity in a complex environment, an evaluation model of regional comprehensive disaster reduction capacity in a complex environment based on remote sensing monitoring and data image feature analysis is proposed. According to the geographical location and scale of disaster spots and the parameter analysis of the model of disaster-bearing bodies around the disaster spots, the remote sensing monitoring method is adopted to extract the geographical remote sensing images of regional disaster spots in a complex environment. The collected geographical remote sensing images of regional disaster points under the complex environmental background are filtered and preprocessed, and the texture parameters of the geographical remote sensing images of regional disaster points under the complex environmental background are recognized by combining the method of image texture feature extraction. Based on the method of tone mapping, the rapid filtering and feature analysis of the geographical remote sensing images of regional disaster points under the complex environmental background are carried out, and the time, position, damage, and so on in the geographical remote sensing images of regional disaster points under the complex environmental background are analyzed. By using the method of parameter analysis and gradient operator operation, a comparison model of geographical remote sensing images of regional disaster points under the complex environmental background is established, and the reliability evaluation of regional comprehensive disaster reduction ability under the complex environmental background is realized according to the method of contrast and detail significance enhancement. The test shows that this method has high accuracy in evaluating regional comprehensive disaster reduction capability under a complex environment, high accuracy in marking the geographical location of regional disaster points under a complex environment, and good fusion performance and reliability of regional comprehensive disaster reduction capability evaluation parameters.

Formation process of thermal damage in a target area of high intensity focused ultrasound and effectiveness analysis of B-ultrasound real-time monitoring

High intensity of focused ultrasound (HIFU) is an effective tumor therapy, taking advantage of the thermal effect and cavitation effect to generate thermal damage to the target tissue. However, inaccurate ultrasonic dose control may result in ineffective or excessive treatment. Thus, real-time monitoring of the thermal damage formation process is critical. To evaluate the effectiveness of real-time monitoring of B-ultrasound, ex-vivo bovine livers were irradiated by 1.155 MHz focused ultrasound with emission time T1 of 200 ms and interval time T2 of 200 ms. For orthogonal experiments, ultrasound was irradiated at sound power of 100 W, 125 W, and 150 W for 10 s, 20 s, and 40 s, respectively. B-ultrasound image sequences are collected using a 7.5 MHz linear array and compared with backscattered echo signals and thermal damage slices, respectively, to build relationships between B-mode ultrasound monitoring and thermal effect or cavitation effect. The experimental results demonstrated that the tissue ablation process caused by thermal effect cannot be effectively monitored using B-ultrasound, but the process caused by the cavitation effect can. The analysis revealed a strong temporal correlation between the appearance of bright spots in B-ultrasound images and the sudden increase of the scattered echo power spectrum, which were caused by a large number of micro-bubbles from cavitation. The damaged cavity structure of the slices and the development trend of micro-bubbles showed a strong spatial correlation. Furthermore, the sudden increase in the scattered echo signal shows the potential of early warning of cavitation, as it is 1.2–2.0 s ahead of the bright spot in the B-ultrasound image, laying the experimental basis for improving the effectiveness of B-ultrasound monitoring in clinical HIFU surgery.

MatGel: A MATLAB program for quantitative analysis of 2D polyacrylamide electrophoresis (2D-PAGE) protein gel images

Two-dimensional polyacrylamide gel electrophoresis (2D-PAGE) is widely used in proteomics studies. Hundreds of proteins extracted from a biological sample can be separated and visualized on a 2D-PAGE gel. The interpretation of protein expression levels relies on the comparison of areas and intensities of the corresponding protein spots in 2D-PAGE gel images. However, determination of protein spot areas by the manual selection method is time-consuming and error-prone. The purpose of this research is to develop a highly automated program for the simultaneous detection and quantification of protein spots across a large number of 2D-PAGE protein gel images by using MATLAB image processing toolbox. This program will enhance the potential of using 2D-PAGE technique as a high throughput quantitative protein expression study tool. We developed MatGel, a simple and efficient program for protein spot area detection and intensity quantification from 2D-PAGE protein gel images. MatGel can detect and output the areas and mean intensities of corresponding protein spots across a large number of 2D-PAGE gel images simultaneously. Users also have options to adjust preferences at each step of image analysis. Basic knowledge with MATLAB programming language is required to run the program.•We developed MATLAB program MatGel to automate the determination of protein spots on 2D-PAGE protein gel images.•MatGel can analyze a large number of 2D-PAGE gel images simultaneously to minimize human errors.•MatGel is flexible and easy to use.

Study on particle generation model of anode in rod-pinch diode

&lt;sec&gt;Flash radiography technology is commonly used in detonation physics experiments and nondestructive testing, in which an X-ray diode is an integral part of flash radiography equipment. Its function is to convert the electric energy stored in the front power supply into X-rays through the bremsstrahlung effect. Rod-pinch diode is one of the most commonly used X-ray diodes in 1–4 MV. It has the characteristics of a small focal spot and high imaging resolution. The anode ions of the rod pinch diode come from the anode plasma, and the anode plasma electrons are generated at the same time as the anode plasma ions. Before the establishment of the bipolar current between the anode and cathode of the rod pinch diode, these electrons are mainly absorbed by the anode; however, after the formation of the bipolar current, due to the zero electric field on the anode surface, plasma electrons will accumulate near the anode surface and will not be absorbed. Given the theoretical derivation of bipolar current in the gap between anode and cathode of rod pinch diode in the early stage, it is necessary to study the electrons in anode plasma.&lt;/sec&gt;&lt;sec&gt;The simulation of the rod-pinch diode is an essential tool for improving the performance of the rod-pinch diode. To improve simulation accuracy, it is necessary to study the emission mechanism of cathode and anode particles and continuously optimize the simulation model. In this paper, PIC and Monte Carlo simulation are used. An anode plasma model is proposed in this work based on the anode ion emission model of the rod-pinch diode and the characteristics of space charge bipolar flow, that is, when the anode plasma environment is fully established, the electric field on the anode surface is zero, and ions and electrons will accumulate on the anode surface. The new model is analyzed in detail and compared with the anode ion emission model in rod-pinch diode current, electromagnetic field distribution between cathode and anode, electron energy spectrum, motion state of electron incident anode rod, dose, and spot size. The results show that the calculation results from the new model are closer to the experimental results, which shows that the role of electrons accumulated near the anode in the plasma cannot be ignored in the numerical calculation of the rod-pinch diode anode rod surface plasma.&lt;/sec&gt;

Cross-Batch Hard Example Mining With Pseudo Large Batch for ID vs. Spot Face Recognition.

In our daily life, a large number of activities require identity verification, e.g., ePassport gates. Most of those verification systems recognize who you are by matching the ID document photo (ID face) to your live face image (spot face). The ID vs. Spot (IvS) face recognition is different from general face recognition where each dataset usually contains a small number of subjects and sufficient images for each subject. In IvS face recognition, the datasets usually contain massive class numbers (million or more) while each class only has two image samples (one ID face and one spot face), which makes it very challenging to train an effective model (e.g., excessive demand on GPU memory if conducting the classification on such massive classes, hardly capture the effective features for bisample data of each identity, etc.). To avoid the excessive demand on GPU memory, a two-stage training method is developed, where we first train the model on the dataset in general face recognition (e.g., MS-Celeb-1M) and then employ the metric learning losses (e.g., triplet and quadruplet losses) to learn the features on IvS data with million classes. To extract more effective features for IvS face recognition, we propose two novel algorithms to enhance the network by selecting harder samples for training. Firstly, a Cross-Batch Hard Example Mining (CB-HEM) is proposed to select the hard triplets from not only the current mini-batch but also past dozens of mini-batches (for convenience, we use batch to denote a mini-batch in the following), which can significantly expand the space of sample selection. Secondly, a Pseudo Large Batch (PLB) is proposed to virtually increase the batch size with a fixed GPU memory. The proposed PLB and CB-HEM can be employed simultaneously to train the network, which dramatically expands the selecting space by hundreds of times, where the very hard sample pairs especially the hard negative pairs can be selected for training to enhance the discriminative capability. Extensive comparative evaluations conducted on multiple IvS benchmarks demonstrate the effectiveness of the proposed method.