Automated Image Analysis System Research Articles

Novel method for coke optical texture measurement and application on coke derived from different ranks of coking coal

AbstractThe optical texture of coke is undoubtedly one of the important factors that determined many properties of coke, but it still mostly depends on the skills and experience of manual analysis. In this paper, a novel automatic measurement method is proposed for coke optical texture. this method through two analyzer angle images(10° and 170°) under polarizing microscope to recognize isochromatic regions which occurs color changed between blue and red by image processing, then characterize the dimensional sizes of isochromatic regions and classify them into different anisotropic texture type. Two fitted ellipse methods are used to characterize the regions' dimensional sizes by the range of shape factorS, followed by classifying different types of anisotropic texture according to the long and short axes of fitted ellipse. An automatic image acquisition and analysis system is established based on the above method. Finally, eight coke samples derived from different ranks of coking coal were selected to test the proposed method. The result shows that it can effectively recognize different types of anisotropic texture and provide composition of different anisotropic texture type for each coke. Moreover, the anisotropic degree of coke shows a significant positive correlation with parent coal rank and a negative correlation with coke reactivity index (CRI).

Skin Lesions Classification Based on Deep Learning Approach

As one kind of skin cancer, melanoma is very dangerous. Dermoscopy-based early detection and reorganization strategy is critical for melanoma therapy. Skin imaging is essential in order to diagnose the tumor type; this had to be accomplished by taking a biopsy from patients. This procedure is very uncomfortable for patients and requires time and highly qualified dermatologists to precisely differentiate between different types of skin cancer. In order to solve these problems, many efforts focused on developing automatic image analysis systems that reduce the suffering of patients, time, and tumor classification error. Here we report a novel strategy based on deep learning technique, which achieved very high skin melanoma diagnosis accuracy. Different pretrained architectures (VGG16, Inception_v3, and ResNet-50) were proposed and applied on 10,135 dermoscopy skin images. A comparison between these models was accomplished with hair removal and nonremoval algorithms. Experimental results showed that the removal of hair had a significant effect on the skin imaging as it helped to improve the training accuracy results. The VGG16 model achieved the highest accuracy (96.9%) with hair removal algorithm, followed by Inception_v3 and ResNet-50 model with an accuracy of 94.2% and 91.8% respectively.

The radiological diagnosis of bronchiectasis: what's in a name?

Diagnosis of bronchiectasis is usually made using chest computed tomography (CT) scan, the current gold standard method. A bronchiectatic airway can show abnormal widening and thickening of its airway wall. In addition, it can show an irregular wall and lack of tapering, and/or can be visible in the periphery of the lung. Its diagnosis is still largely expert based. More recently, it has become clear that airway dimensions on CT and therefore the diagnosis of bronchiectasis are highly dependent on lung volume. Hence, control of lung volume is required during CT acquisition to standardise the evaluation of airways. Automated image analysis systems are in development for the objective analysis of airway dimensions and for the diagnosis of bronchiectasis. To use these systems, clear and objective definitions for the diagnosis of bronchiectasis are needed. Furthermore, the use of these systems requires standardisation of CT protocols and of lung volume during chest CT acquisition. In addition, sex- and age-specific reference values are needed for image analysis outcome parameters. This review focusses on today's issues relating to the radiological diagnosis of bronchiectasis using state-of-the-art CT imaging techniques.

Cross-evaluation of five slidemakers and three automated image analysis systems: The pitfalls of automation?

Automated slidemakers and stainers and digital microscopes are coupled with haematology analysers to achieve better efficiency and cost-effectiveness. This study evaluates the integrated performance of slidemakers and digital microscopes commonly available on the market. We compared the percentage of neutrophils for five slidemakers (two Siemens Advia Autoslides, a SysmexSP-10 and SP-50 and an Abbot Alinity hs) and a Horiba Hemaprep to the corresponding haematology analyser data (Siemens Advia 2120i, Sysmex XN and Abbot Alinity hq). Differential leucocyte counting (DLC) was performed on three different CellaVision digital microscopes (DM96, DM1200 and DI-60) and manually. The quality of the smears was assessed using a CellaVision SmearChecker. We observed a significant positive absolute bias (P<.05) for the percentage of neutrophils with the Autoslide and Alinity hs smears on the digital microscopes, but not when DLC was performed manually. The SP-10 and SP-50 showed no bias regardless of the DLC method. No bias was observed for the Hemaprep smears. All the smears had an acceptable monolayer quality, stain intensity and colour. All smears, except those from Sp-10, were of an acceptable length. Users should be aware of a potential lack of accuracy that can be encountered when using some slidemakers and digital microscopes. All laboratories should validate or verify the differential counts from slidemakers and digital microscope with automated cell differential counters. Manual count validation should only be considered if a significant proportion of clinically relevant abnormal cells are present. Otherwise, haematology analyser results should be favoured.

Angiogenesis in mammary Paget disease: histopathological analyses of blood vessel density and angiogenic factors

BackgroundWe examined the vascularity of mammary Paget disease histologically to confirm the increased blood flow observed previously by clinical imaging. The relationships among blood vessel density (BVD), histopathological parameters of blood flow in the nipple, and the expression of angiogenic factors such as basic fibroblast growth factor (bFGF) and vascular endothelial growth factor A (VEGFA) were examined.MethodsWe calculated the average CD34-positive BVD and podoplanin (D2–40)-positive lymphatic vessel density (LVD) and the proportion of proliferating of endothelial cells in 14 Paget disease, 3 dermatitis biopsy, and 14 age-matched control cases. As a parameter related to blood flow in the nipple, the total CD34-positive blood vessel lumen area relative to the entire nipple area was measured in each Paget disease and control case using an automated image analysis system. Immunohistochemical expression of bFGF and VEGFA in Paget cells was also examined.ResultsThe average BVD and LVD were significantly higher in the Paget disease cases than in the dermatitis (p = 0.003) and control (p < 0.001) cases. The proportion of proliferating endothelial cells was also increased in the Paget disease cases. The ratio of the CD34-positive blood vessel lumen area to nipple area was also significantly higher in the Paget disease than control cases (p = 0.003). The average BVD was correlated with the average LVD (r = 0.734, p < 0.001) and ratio of the blood vessel lumen area to nipple area (r = 0.692, p < 0.001). Immunohistochemical expression of bFGF was strong in all Paget disease cases, with a significantly higher expression score in the Paget disease than dermatitis (p = 0.003) and control (p < 0.001) cases. The bFGF, but not VEGFA, expression score, was strongly correlated with the average BVD (r = 0.818, p < 0.001) and ratio of the blood vessel lumen area to nipple area (r = 0.503, p = 0.006).ConclusionThese results provide direct histopathological evidence of a marked increase in nipple blood flow in Paget disease detected by clinical imaging. bFGF is considered to play a pivotal role in angiogenesis in mammary Paget disease.

Method for Determining the Measurement Uncertainty of the Detailing Coefficients of the Wavelet Transform of Image Brightness Profiles

This study analyzed the relevance of using measurement methods based on visual data and identified the main problems that arise when calculating the measurement uncertainty of the estimates of quantitative characteristics of images in automated image analysis systems due to the lack of general recommendations for this type of calculation. Moreover, the study highlights the challenges in developing a method for calculating the measurement uncertainty of the detailing coefficients of the wavelet transform applied to the automated analysis of gas-discharge radiation images of water. To evaluate the biological properties of water and highlight the corresponding informative features of the gas-discharge radiation images, a one-dimensional wavelet transform of the brightness profile is proposed. This method of calculating the uncertainty is based on the results of repeated cases and helps reveal the measurement uncertainty of the quantitative characteristics of the images.

Automated image analysis system for studying cardiotoxicity in human pluripotent stem cell-Derived cardiomyocytes

BackgroundCardiotoxicity, characterized by severe cardiac dysfunction, is a major problem in patients treated with different classes of anticancer drugs. Development of predictable human-based models and assays for drug screening are crucial for preventing potential drug-induced adverse effects. Current animal in vivo models and cell lines are not always adequate to represent human biology. Alternatively, human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) show great potential for disease modelling and drug-induced toxicity screenings. Fully automated high-throughput screening of drug toxicity on hiPSC-CMs by fluorescence image analysis is, however, very challenging, due to clustered cell growth patterns and strong intracellular and intercellular variation in the expression of fluorescent markers.ResultsIn this paper, we report on the development of a fully automated image analysis system for quantification of cardiotoxic phenotypes from hiPSC-CMs that are treated with various concentrations of anticancer drugs doxorubicin or crizotinib. This high-throughput system relies on single-cell segmentation by nuclear signal extraction, fuzzy C-mean clustering of cardiac α-actinin signal, and finally nuclear signal propagation. When compared to manual segmentation, it generates precision and recall scores of 0.81 and 0.93, respectively.ConclusionsOur results show that our fully automated image analysis system can reliably segment cardiomyocytes even with heterogeneous α-actinin signals.

Quantitative and qualitative evaluation of deep learning automatic segmentations of corneal endothelial cell images of reduced image quality obtained following cornea transplant.

We are developing automated analysis of corneal-endothelial-cell-layer, specular microscopic images so as to determine quantitative biomarkers indicative of corneal health following corneal transplantation. Especially on these images of varying quality, commercial automated image analysis systems can give inaccurate results, and manual methods are very labor intensive. We have developed a method to automatically segment endothelial cells with a process that included image flattening, U-Net deep learning, and postprocessing to create individual cell segmentations. We used 130 corneal endothelial cell images following one type of corneal transplantation (Descemet stripping automated endothelial keratoplasty) with expert-reader annotated cell borders. We obtained very good pixelwise segmentation performance (e.g., Dice , , across 10 folds). The automated method segmented cells left unmarked by analysts and sometimes segmented cells differently than analysts (e.g., one cell was split or two cells were merged). A clinically informative visual analysis of the held-out test set showed that 92% of cells within manually labeled regions were acceptably segmented and that, as compared to manual segmentation, automation added 21% more correctly segmented cells. We speculate that automation could reduce 15 to 30min of manual segmentation to 3 to 5min of manual review and editing.

A feasible evaluation protocol to determine the most reactive sulfide-bearing aggregate for use in concrete

Internal sulfate attack is one of the most harmful deterioration mechanisms to concrete. A testing procedure was developed to evaluate how reactive is a sulfide-bearing aggregate to use in concrete. The aggregates were produced from iron sulfide rocks that have at least 1.5% of sulfur. Particle sizes between 0.15- and 4.80-mm had the sulfur quantities from sulfide (pyrite and pyrrhotite) determined by direct combustion test. The exposed surface of each sulfide was calculated using a scanning electron microscope (SEM) connected to an automated image analysis system - Mineral Liberation Analyzer (MLA). Samples also were immersed in an H2O2 solution and the released sulfate (from the sulfide oxidation) was quantified. The results showed that the larger the sulfide exposed surface and the smaller the particle size are, the more effective is its oxidation and consequently more reactive the aggregate.

The Effect of Fly Ash Microspheres on the Pore Structure of Concrete

The fly ash microspheres (FAMs) formed during the mineral transformation stage in coal combustion are hollow spherical particles with a density less than water. This paper presents the results of X-ray micro-computed tomography and an automatic image analysis system of the porosity in the structure of hardened concrete with microspheres. Concrete mixtures with ordinary Portland cement and two substitution rates of cement by microspheres—5% and 10%—are investigated. For all considered mixes, a constant water/binder ratio (w/b) equal to 0.50 was used. The distribution of the air voids and the compressive strength of the concrete were tested after 28 days. With the increasing mass of cement replacement by FAMs, the compressive strength decreases after 28 days. The total volume of the air voids in hardened concrete with fly ash microspheres tested by X-ray varies from 5.1% to 7.4%. The closed pores constitute more than 80% of the total content of air pores. The study proves that the use of microspheres grains with specific dimensions has a significant impact on concrete porosity. Their application in concrete technology can be an alternative aeration solution for fresh concrete mixes and an effective method for utilization.

Method calculating of uncertainty in evaluating the detailing coeffi cients of wavelet-conversion of profi le brightness image

The analysis of the relevance of the use of measurement methods based on obtaining visual data. The main problems that arise in automated image analysis systems in calculating the uncertainty of estimates of quantitative characteristics of images are identifi ed. Since modern methods of collecting and transforming visual data are characterized by great diversity, at the moment there are no general recommendations for calculating the uncertainty of estimates of their geometric and brightness parameters. The article is devoted to the development of a method for calculating uncertainty in evaluating the detailed coeffi cients of the wavelet transform, which is used in the automated analysis of images of gas-discharge radiation of water. The method based on recording gas-discharge radiation of water samples in an external pulsed electromagnetic fi eld was chosen to study the biological properties of water, since the formation of a gas discharge in the air gap between conductors directly depends on the presence of free charge carriers that can be simulated from the surface of the object under study. The number of free charge carriers determines the biological properties of water, since they affect the ability of drinking water to provide metabolic processes in living organisms by maintaining the normal course of redox reactions at the cellular level. To assess the biological properties of water and highlight the corresponding informative features of gas-discharge images, it is proposed to use the method of one-dimensional wavelet transform of the brightness profi le. A method for calculating the uncertainty in evaluating the detailed coeffi cients of the wavelet transform is considered. The proposed method for calculating the uncertainty is based on the results of repeated observations and allows us to estimate the uncertainty of the measurement results presented in the form of quantitative signs of images.

Automated image analysis system for renal filtration barrier integrity of potassium bromate treated adult male albino rat

Potassium bromate (KBrO3) is a potent nephrotoxic agent that leads to a significant decrease in the activities of renal antioxidant capacity, antioxidant loss and restoration of the renal dysfunction. Several measurements are used to examine the kidney status, including the base width of the foot, the slit pore diameter, and the glomerular basement membrane thickness of the kidney. In this work, morphometric analysis based on image processing is carried out to assess the filtration barrier integrity parameters, which indicates the degree of recovery against the nephrotoxic effect of the KBrO3 on the renal cortex of adult male albino rat and assesses the capability of the renal cortex to recover after its cessation. The morphometric methods based proposed image analysis system enabled the identification of the renal status of different groups, namely the control, potassium bromate affected, and the recovered groups, according to the variation of the measured parameters is a powerful tool. The proposed image analysis system provided a radical geometric morphometrics, which includes morphological operations and structuring element processes in order to identify the glomerular filtration barrier and the feet for further measurements in each case study. The results established that the average lengths of the feet in the histological microscopic images are 465.2397 nm, 278.189 nm, and 393.2347 nm for the control, KBrO3 affected rats and the recovered rats; respectively.

Analysis of Wear Resistance Characteristics of Additive Products Made from Electroerosive Cobalt-Chromium Powders

Purpose of research of this research is to study wear resistance characteristics of additive products made from electroerosive cobalt-chromium powders. Methods . Wastes of cobalt-chromium CELLIT alloy were used for the research [27]. Butyl alcohol was used as a working fluid. A unit for layer-by-layer application of powder materials with plasma was used for experimental samples production. The intensity and wear rate of the sample surface and counterface were measured on Tribometer automated friction machine, CSM Instruments. The sample was mounted in a holder. A rod was fixed perpendicular to the surface of a sample. A 6 mm diameter ball made of Stainless Steel AISI 420 (Vickers hardness 5000-8000 HV) was at the end of the rod. Radius of wear curvature was selected by adjusting the displacement sensor. Another sensor compensated the friction force and allowed to set the friction coefficient value at a certain time. The condition of friction surfaces was studied with the help of optical inverted microscope Olympus GX 51 equipped with SIMAGIS Photolab automated image analysis system and Quanta 200 3D electron ion scanning microscope. Results . According to the performed analysis of wear resistance characteristics of additive products made from electroerosive cobalt-chromium powders friction coefficient varies from 0.144 to 0.602. An average friction coefficient is 0.526. A wear factor of the sample is two times higher than the wear factor of counterface. Conclusion . Performed research allows expanding the information about properties of additive products made from electroerosive cobalt-chromium powders and reducing production cost.

A Novel Automated Image Analysis System Using Deep Convolutional Neural Networks to Diagnose MDS

Detection of dysmorphic cells in peripheral blood (PB) smears is essential for diagnosis of hematological malignancies. Myelodysplastic syndromes (MDS) are heterogeneous hematopoietic stem cell disorders that can lead to acute leukemia. Although examinations of bone marrow aspiration and biopsy as well as chromosomal and genetic tests are essential to diagnose these disorders, conventional tests such as complete blood count (CBC) and peripheral blood (PB) smear examinations remain to be initial diagnostic work ups. Detection of dysplastic cells in PB smears and evaluation of CBCs are particularly useful for rapid screening. Thanks to the recent advancement of computational and laboratory technologies, routine manual microscopic examinations have been replaced by automated hematology analyzers in many hematology laboratories. However, detection of dysmorphic cells is still challenging. Therefore, more sophisticated image recognition systems need to be developed. In this study, we developed a novel MDS diagnostic system using PB smears. The system consists of a convolutional neural network (CNN)-based image recognition deep learning system (DLS) and an EGB-based decision-making algorithm (XGBoost). All PB smears were prepared at Juntendo University Hospital. The slides were stained with May Grunwald-Giemsa using a fully automated slide-maker. A total of 703,970 digitalized cell images were collected. First, we trained the CNN-based image-recognition system using 695,030 blood cell images taken from 3,261 PB smears of which 1,165 were obtained from patients with hematological disorders. The hematological disorders included MDS (n=94 cases), myeloproliferative neoplasms (n=127), acute myeloid leukemia (n= 38), acute lymphoblastic leukemia (ALL, n=27), malignant lymphoma (n=324), multiple myeloma (n=82) and AA (n=42). Of all images, these 695,030 images were used to train the CNN-based image-recognition system (Fig 1), and rest of the images (n=8,940) were used for validation. The datasets were classified into 17 cell types and 97 abnormal morphological features by two board-certified laboratory technologists and one senior hematopathologist using the morphological criteria of the Clinical and Laboratory Standards Institute (CLSI) H20-A2 guideline and the 2017 WHO classification of myeloid neoplasms and acute leukemia. After accumulating the image patterns using the training datasets, the performance of the DLS was evaluated using the validation datasets that were prepared for testing the DLS. The internal features learnt by the DLS using t-distributed Stochastic Neighbor Embedding (t-SNE)(Fig 2). In this context, our CNN-based image-recognition system exhibited a sensitivity of &gt;93.5% and a specificity of &gt;96.0% when classifying cells in subsets of the validation datasets. We then created an automated MDS diagnostic system by combining the CNN-based image-recognition system with a form of XGBoost. To establish diagnostic algorithm, the training datasets obtained from 75 MDS and 36 Aplastic Anemia (AA) cases were used for learning of the cell image pattern for each disease. The diagnosis of all datasets was validated by independent hematopathologists using clinical information, laboratory, flow cytometric, and genetic data, and bone marrow aspiration and biopsy findings. The accuracy of the system was tested using validation datasets (26 MDS and 11 AA cases). The system differentiated MDS from AA with the sensitivity and specificity of 96.2% and 100%, respectively (AUC 0.990). In conclusion, this is the first CNN-based automated initial diagnostic system for MDS using PB smears, which is applicable to develop new automated image diagnostic systems for various hematological disorders. Currently, we are collecting more data to improve the accuracy. Disclosures Kimura: Sysmex Corporation: Employment. Takehara:Sysmex Corporation: Employment. Uchihashi:Sysmex Corporation: Employment.

SINGLE-IMAGE DEHAZING ON AERIAL IMAGERY USING CONVOLUTIONAL NEURAL NETWORKS

Abstract. Haze contains floating particles in the air which can result in image quality degradation and visibility reduction in airborne data. Haze removal task has several applications in image enhancement and can improve the performance of automatic image analysis systems, namely object detection and segmentation. Unlike rich haze removal literature in ground imagery, there is a lack of methods specifically designed for aerial imagery, considering the fact that there is a characteristic difference between the aerial imagery domain and ground one. In this paper, we propose a method to dehaze aerial images using Convolutional Neural Networks (CNNs). Currently, there is no available data for dehazing methods in aerial imagery. To address this issue, we have created a syntheticallyhazed aerial image dataset to train the neural network on aerial hazy image dataset. We train All-in-One dehazing network (AODNet) as the base approach on hazy aerial images and compare the performance of our proposed approach against the classical model. We have tested our model on natural as well as the synthetically-hazed aerial images. Both qualitative and quantitative results of the adapted network show an improvement in dehazing results. We show that the adapted AOD-Net on our aerial image test set increases PSNR and SSim by 2.2% and 9%, respectively.

Automated image analysis techniques to characterise pulverised coal particles and predict combustion char morphology

A new automated image analysis system that analyses individual coal particles to predict daughter char morphology is presented. 12 different coals were milled to 75–106 µm, segmented from large mosaic images and the proportions of the different petrographic features were obtained from reflectance histograms via an automated Matlab system. Each sample was then analysed on a particle by particle basis, and daughter char morphologies were automatically predicted using a decision tree-based system built into the program. Predicted morphologies were then compared to ‘real’ char intermediates generated at 1300 °C in a drop-tube furnace (DTF). For the majority of the samples, automated coal particle characterisation and char morphology prediction differed from manually obtained results by a maximum of 9%. This automated system is a step towards eliminating the inherent variability and repeatability issues of manually operated systems in both coal and char analysis. By analysing large numbers of coal particles, the char morphology prediction could potentially be used as a more accurate and reliable method of predicting fuel performance for power generators.

A novel automated image analysis system using deep convolutional neural networks can assist to differentiate MDS and AA

Detection of dysmorphic cells in peripheral blood (PB) smears is essential in diagnostic screening of hematological diseases. Myelodysplastic syndromes (MDS) are hematopoietic neoplasms characterized by dysplastic and ineffective hematopoiesis, which diagnosis is mainly based on morphological findings of PB and bone marrow. We developed an automated diagnostic support system of MDS by combining an automated blood cell image-recognition system using a deep learning system (DLS) powered by convolutional neural networks (CNNs) with a decision-making system using extreme gradient boosting (XGBoost). The DLS of blood cell image-recognition has been trained using datasets consisting of 695,030 blood cell images taken from 3,261 PB smears including hematopoietic malignancies. The DLS simultaneously classified 17 blood cell types and 97 morphological features of such cells with >93.5% sensitivity and >96.0% specificity. The automated MDS diagnostic system successfully differentiated MDS from aplastic anemia (AA) with high accuracy; 96.2% of sensitivity and 100% of specificity (AUC 0.990). This is the first CNN-based automated initial diagnostic system for MDS using PB smears, which is applicable to develop new automated diagnostic systems for various hematological disorders.

Automatic image analysis system to measure wound area in vitro

Abstract In-vitro wound area measurement tracks the rate of wound healing. This project develops and validates an automatic image analysis system to calculate wound area from digital images of an in-vitro 3D tissue model wounded with a biopsy punch. The algorithms were evaluated for repeatability, reliability, and reproducibility, and validated against a known area. Repeatability was checked through repeated measurements under repeated conditions. Reproducibility was evaluated using a Bland Altman plot and paired t-test. Reliability was validated using an image of a known pixel area as a control. The validated image analysis system then calculated wound area from the digital camera and microscope images obtained from an in vitro photo biomodulation treatment experiment. A total of 48 wounded tissues were grouped into red and blue light treatment groups and untreated controls. All daily images were fed into the image analysis system to calculate wound area. The wound area (normalized by day 0) is plotted along the 2-week treatment experiment period to observe wound area in time. The normalised wound area plotted across treatment days show no change in wound area during the treatment period. Future work will adapt the imaging system for visualizing the reepithelialisation cell front marked by live dyes.

Rapid Processing of Perfusion and Diffusion for Ischemic Strokes in the Extended Time Window: An Indian Experience.

Background:Endovascular treatment of acute ischemic stroke with large-vessel occlusion is the standard of care now. Initially restricted to 6 h after onset, the treatment can now be offered to selected patients up to 24 h based on clinical and imaging criteria.Objective:Perfusion imaging can help in identifying patients who may benefit from endovascular treatment in the extended time window. Manual analysis of perfusion images is time and skill intensive. Rapid processing of perfusion and diffusion (RAPID) is an automated image analysis system that analyzes perfusion maps. We report our initial experience of using this system in selection of patients for endovascular stroke treatment.Methods:All patients who presented with acute stroke underwent baseline imaging with computed tomography (CT) and CT angiogram or magnetic resonance imaging (MRI) and MR angiogram. Patients presenting between 6 and 24 h after onset underwent perfusion imaging, which was analyzed by RAPID. The results were used to select the patients who then underwent mechanical thrombectomy.Results:RAPID results identifying ischemic core and hypoperfused tissue were available within 5 min in each of the three cases. At 3 months, all patients showed improvement in the modified Rankin Scale.Conclusion:In extended time windows, RAPID provides a fast and reliable estimate of salvageable brain tissue to help select patients for endovascular treatment.

Computerized Ultrasonic Imaging Inspection: From Shallow to Deep Learning.

For many decades, ultrasonic imaging inspection has been adopted as a principal method to detect multiple defects, e.g., void and corrosion. However, the data interpretation relies on an inspector’s subjective judgment, thus making the results vulnerable to human error. Nowadays, advanced computer vision techniques reveal new perspectives on the high-level visual understanding of universal tasks. This research aims to develop an efficient automatic ultrasonic image analysis system for nondestructive testing (NDT) using the latest visual information processing technique. To this end, we first established an ultrasonic inspection image dataset containing 6849 ultrasonic scan images with full defect/no-defect annotations. Using the dataset, we performed a comprehensive experimental comparison of various computer vision techniques, including both conventional methods using hand-crafted visual features and the most recent convolutional neural networks (CNN) which generate multiple-layer stacking for representation learning. In the computer vision community, the two groups are referred to as shallow and deep learning, respectively. Experimental results make it clear that the deep learning-enabled system outperformed conventional (shallow) learning schemes by a large margin. We believe this benchmarking could be used as a reference for similar research dealing with automatic defect detection in ultrasonic imaging inspection.