VGG16 Network Research Articles

Abstract P-5: Neural Network Approaches to Classify 3D Protein Structures from the Data of X-ray Laser Radiation Diffraction from Single Particles

Background: Protein structure determination using X-ray free-electron laser (XFEL) includes analysis and merging a large number of snapshot diffraction patterns. Convolutional neural networks are widely used to solve numerous computer vision problems, e.g. image classification, and can be used for diffraction pattern analysis. But the task of protein structure determination with the use of CNNs only is not yet solved. Methods: We simulated the diffraction patterns using the Condor software library and obtained more than 1000 diffraction patterns for each structure with simulation parameters resembling real ones. To classify diffraction patterns, we tried two approaches, which are widely known in the area of image classification: a classic VGG network and residual networks. Results: 1. Recognition of a protein class (GPCRs vs globins). Globins and GPCR-like proteins are typical α-helical proteins. Each of these protein families has a large number of representatives (including those with known structure) but we used only 8 structures from every family. 12,000 of diffraction patterns were used for training and 4,000 patterns for testing. Results indicate that all considered networks are able to recognize the protein family type with high accuracy. 2. Recognition of the number of protein molecules in the liposome. We considered the usage of lyposomes as carriers of membrane or globular proteins for sample delivery in XFEL experiments in order to improve the X-ray beam hit rate. Three sets of diffractograms for liposomes of various radius were calculated, including diffractograms for empty liposomes, liposomes loaded with 5 bacteriorhodopsin molecules, and liposomes loaded with 10 bacteriorhodopsin molecules. The training set consisted of 23625 diffraction patterns, and test set of 7875 patterns. We found that all networks used in our study were able to identify the number of protein molecules in liposomes independent of the liposome radius. Our findings make this approach rather promising for the usage of liposomes as protein carriers in XFEL experiments. Conclusion: Thus, the performed numerical experiments show that the use of neural network algorithms for the recognition of diffraction images from single macromolecular particles makes it possible to determine changes in the structure at the angstrom scale.

Research on Super-Resolution Reconstruction Algorithm of Image Based on Generative Adversarial Network

Image super-resolution is to use a series of algorithms to improve the original image resolution. The process of obtaining high-resolution image through some low-resolution images is image super-resolution reconstruction technology. There are two main research fields in super-resolution reconstruction, one is to restore the real details of the image, and the other is to not require too much detail, and only pay attention to the overall visual effect of the image. In this paper, an improved super-resolution reconstruction algorithm based on generative adversarial network is proposed. The network model and loss function are improved and optimized. The auxiliary VGG-19 network is used to extract the image features, and the extended convolution is used to expand the receptive field, which makes the image have a better reconstruction effect. Using DIV2k data set as training set and set5, set14, bsd100 data sets as test set, a series of experimental analysis is to prove the feasibility of the procedure method. Compared with the existing mainstream models, the perceptual effect of the image has been improved.

Multi-Scale Object Detection Method Based on Multi-Branch Parallel Dilated Convolution

<p indent=0mm>Existing object detection algorithms only use a fixed size convolution kernel when extracting features, ignoring the difference in the receptive field of different scale features, which affects the detection effect of different scale objects. To solve this problem, a multi-scale object detection network based on multi-branch parallel dilated convolution is proposed. Firstly, the basic network VGG-16 is used to extract the features of the image. Secondly, a multi-branch parallel dilated convolution is designed to extract multi-scale features to improve object detection ability of the network. Then, a non-local block is employed to integrate the global spatial information and enhance the context information. Finally, the object detection and location tasks are performed on feature maps with different scales. Experimental results on PASCAL VOC and MS COCO datasets demonstrate that the proposed method can effectively improve the detection accuracy of different scale objects and clearly improve the detection accuracy of small objects.

Multi-feature fusion tracking algorithm based on generative compression network

In order to solve the problem of inaccurate positioning in the process of target tracking due to illumination, scale change and occlusion, a correlation filtering tracking algorithm based on generate compression network is proposed. Firstly, the encoder is used to compress the high-dimensional depth features extracted from VGG16 network, and the soft quantizer reduces floating-point operation so that improves the operation speed. Secondly, the discriminator of generative adversarial network guides the encoder and generator to better compress and recover the original depth features by generating a discriminator against the network, so as to enhance the ability of the encoder to extract the key features of the target, then, the compressed depth features, gray features and hog features are combined to improve the ability of target representation. Thirdly, the correlation filter and PCA feature dimensionality reduction are used to complete the target precise location and discriminant scale estimation. The experimental results show that the tracking accuracy of the proposed algorithm is 87.1%, and the tracking rate is up to 70fps.

Moving Objects Semantic Segmentation using SegNet with VGG Encoder for Autonomous Driving

Segmentation and recognition become the general steps to identify objects. This research discusses pixel-wise semantic segmentation based on moving objects. The data from the CamVid video which is a collection of autonomous driving images. The image data consist of 701 images accompanied by labels. The segmentation and recognition of 11 objects contained in the image (sky, building, pole, road, pavement, tree, sign-symbol, fence, car, pedestrian and bicyclist) is representing. This moving object segmentation is carried out using SegNet which is one of the Convolutional Neural Network (CNN) methods. Image segmentation on CNN generally consists of two parts: Encoder and Decoder. VGG16 and VGG19 pre-trained networks are used as encoders, while decoders are the upsampling of encoders. Network optimization uses stochastic gradient descent of Momentum (SGDM). The test produces the best recognition was road objects with an accuracy of 0.96013, IoU 0.93745, F1-Score 0.8535 using VGG19 encoder, while when using VGG16 encoder accuracy was 0.94162, IoU 0.92309, and F1-Score 0.8535.

SCCDNet: A Pixel-Level Crack Segmentation Network

Cracks are one of the most serious defects that threaten the safety of bridges. In order to detect different forms of cracks in different collection environments quickly and accurately, we proposed a pixel-level crack segmentation network based on convolutional neural networks, which is called the Skip Connected Crack Detection Network (SCCDNet). The network is composed of three parts: the Encoder module with 13 convolutional layers pretrained in the VGG-16 network, the Decoder module with a densely connected structure, and the Skip-Squeeze-and-Excitation (SSE) module which connects the feature map shaving the same resolution in the Encoder and Decoder. We used depthwise separable convolution to improve the accuracy of crack segmentation while reducing the complexity of the model. In this paper, a dataset containing cracks collected in different scenes was established, and SCCDNet was trained and tested on this dataset. Compared with the advanced models, SCCDNet obtained the best crack segmentation performance, while F-score reached 0.7763.

Detection and parameter estimation of frequency hopping signal based on the deep neural network

ABSTRACT In order to solve the problem of low detection rate of frequency hopping (FH) signal under low SNR, an FH signal detection technology based on deep neural network (vgg16 network) is proposed. Firstly, this paper introduces the principle of FH signal. Then, this paper presents the SSD deep neural network detection framework, and uses this framework to detect FH signal and estimate its parameters. Time-frequency distribution maps of FH signals are used as the input of deep neural network model to carry out tag learning and training, and then the trained model is used to realise FH signal detection. At the same time, aiming at the noise problem of time-frequency distribution map, the time-frequency graph correction method of K-means clustering algorithm is proposed. After modification, it is more tolerant to noise. The simulation results show that when the SNR is –4 dB, the detection rate of FH signal can reach more than 88%. Finally, the performance of parameter estimation of FH signal is analysed.

Novel Deep Learning Radiomics Model for Preoperative Evaluation of Hepatocellular Carcinoma Differentiation Based on Computed Tomography Data

Background: The evaluation of tumor differentiation is an urgent clinical issue that would facilitate the establishment of individualized therapeutic strategies. Our aims were to develop a deep learning radiomics model based on computed tomography (CT) data for preoperative evaluation of hepatocellular carcinoma (HCC) differentiation (low vs high grade), and to preliminarily explore the biological basis of the radiomics model. Methods: A total of 1234 HCC patients with contrast-enhanced CT images were recruited from two institutions. Radiomics features were extracted from preoperative venous-phase CT data and selected in terms of reproducibility, relevance, and redundancy. The random forest (RF) algorithm was applied to establish a radiomics signature. Meanwhile, the deep learning model was constructed based on a VGG network. Additionally, clinical characteristics of the subjects were used to construct a clinical model. Finally, the fused model integrated a tripartite prediction based on a logistic regression algorithm. A correlation analysis was performed to explore the association between radiomics features and biological variables based on multiomics levels. Findings: The radiomics signature established with the RF algorithm comprised 25 radiomics features. The AUCs in the training, internal validation, and independent test cohorts were 0·82, 0·76, and 0·75, respectively, for the radiomics signature; 0·85, 0·81, and 0·75, respectively, for the deep learning model; and 0·89, 0·83, and 0·80, respectively, for the fused model. Multiomics analysis showed that the selected radiomics features contained abundant biological information that was related to tumor differentiation. Interpretation: Our deep learning radiomics model can serve as a noninvasive tool of preoperative HCC differentiation evaluation to guide clinical decision-making and prognostic stratification. Funding: National Key Research and Development Program of China, Natural Science Foundation of China, Zhejiang Provincial Education Department. Declaration of Interest: The authors declare that they have no competing interests. Ethical Approval: This study was approved by the institutional review boards of the First Affiliated Hospital, College of Medicine, Zhejiang University (Institution 1) and Ningbo Medical Center Lihuili Hospital (Institution 2), which waived the requirement for patients’ informed consent.

Deep Learning Based Real Age and Gender Estimation from Unconstrained Face Image towards Smart Store Customer Relationship Management

The COVID-19 pandemic markedly changed the human shopping nature, necessitating a contactless shopping system to curb the spread of the contagious disease efficiently. Consequently, a customer opts for a store where it is possible to avoid physical contacts and shorten the shopping process with extended services such as personalized product recommendations. Automatic age and gender estimation of a customer in a smart store strongly benefit the consumer by providing personalized advertisement and product recommendation; similarly, it aids the smart store proprietor to promote sales and develop an inventory perpetually for the future retail. In our paper, we propose a deep learning-founded enterprise solution for smart store customer relationship management (CRM), which allows us to predict the age and gender from a customer’s face image taken in an unconstrained environment to facilitate the smart store’s extended services, as it is expected for a modern venture. For the age estimation problem, we mitigate the data sparsity problem of the large public IMDB-WIKI dataset by image enhancement from another dataset and perform data augmentation as required. We handle our classification tasks utilizing an empirically leading pre-trained convolutional neural network (CNN), the VGG-16 network, and incorporate batch normalization. Especially, the age estimation task is posed as a deep classification problem followed by a multinomial logistic regression first-moment refinement. We validate our system for two standard benchmarks, one for each task, and demonstrate state-of-the-art performance for both real age and gender estimation.

Areal‐timeclassification and detection method for mutton parts based on single shot multi‐box detector

AbstractThis article proposes a real‐time classification and detection method for mutton parts based on a single shot detector (SSD). We acquired 9,000 images of various parts of mutton in a sheep slaughtering workshop, characterized by multiple classes and multiple samples. After image preprocessing, an image dataset of the mutton parts was established for later model training. Subsequently, we introduced transfer learning to train an SSD–VGG network and obtain the optimal model. The optimal model was then applied to determine the category and position of each mutton part in the image, thus realizing the classification and detection of mutton parts. In this method, the average accuracymAPand average processing time for a single image are selected as the accuracy and speed indicators, respectively, for judging the detection performance of the model. The feature extraction network VGG is replaced with MobileNetV1 to optimize the real‐time performance of the SSD. Furthermore, we set an additional illumination dataset with two brightness levels “bright” and “dark” to verify the generalization ability of the optimized model. Finally, four common object detection algorithms, namely YoloV3–MobileNetV1, YoloV3–DarkNet53, Fast‐RCNN, and Cascade‐RCNN, are introduced to perform comparative experiments on mutton image datasets. The test results prove that the SSD–MobileNetV1 exhibits high accuracy and good real‐time performance, with a certain generalization ability. It has a better comprehensive detection ability than other methods and can provide technical support for mutton processing.Practical ApplicationsCurrently, in the processing of mutton, the multiple parts of mutton are identified and sorted manually, which is time‐consuming and laborious, and there are certain hidden food safety hazards. A deep‐learning‐based object detection method can solve the above problems effectively. Therefore, this study uses SSD to perform an accurate real‐time recognition of the multiple parts of mutton from its images and provide a visual guidance for mutton sorting robots. It can also aid further research in the slaughtering and processing of other meat.

Prediction of soft tissue sarcoma response to radiotherapy using longitudinal diffusion MRI and a deep neural network with generative adversarial network-based data augmentation.

The goal of this study was to predict soft tissue sarcoma response to radiotherapy (RT) using longitudinal diffusion-weighted MRI (DWI). A novel deep-learning prediction framework along with generative adversarial network (GAN)-based data augmentation was investigated for the response prediction. Thirty soft tissue sarcoma patients who were treated with five-fraction hypofractionated radiation therapy (RT, 6Gy×5) underwent diffusion-weighted MRI three times throughout the RT course using an MR-guided radiotherapy system. Pathologic treatment effect (TE) scores, ranging from 0-100%, were obtained from the post-RT surgical specimen as a surrogate of patient treatment response. Patients were divided into three classes based on the TE score (TE≤20%, 20%<TE<90%, TE≥90%). Apparent diffusion coefficient (ADC) maps of the tumor from the three time points were combined as 3-channel images. An auxiliary classifier generative adversarial network (ACGAN) was trained on 20 patients to augment the data size. A total of 15,000 synthetic images were generated for each class. A prediction model based on a previously described VGG-19 network was trained using the synthesized data, validated on five unseen validation patients, and tested on the remaining five test patients. The entire process was repeated seven times, each time shuffling the training, validation, and testing datasets such that each patient was tested at least once during the independent test stage. Prediction performance for slice-based prediction and patient-based prediction was evaluated. The average training and validation accuracies were 86.5%±1.6% and 84.8%±1.8%, respectively, indicating that the generated samples were good representations of the original patient data. Among the seven rounds of testing, slice by slice prediction accuracy ranged from 81.6% to 86.8%. The overall accuracy of the independent test sets was 83.3%. For patient-based prediction, 80% was achieved in one round and 100% was achieved in the remaining six rounds. The mean accuracy was 97.1%. This study demonstrated the potential to use deep learning to predict the pathologic treatment effect from longitudinal DWI. Accuracies of 83.3% and 97.1% were achieved on independent test sets for slice-based and patient-based prediction respectively.

EMDS-5: Environmental Microorganism image dataset Fifth Version for multiple image analysis tasks.

Environmental Microorganism Data Set Fifth Version (EMDS-5) is a microscopic image dataset including original Environmental Microorganism (EM) images and two sets of Ground Truth (GT) images. The GT image sets include a single-object GT image set and a multi-object GT image set. EMDS-5 has 21 types of EMs, each of which contains 20 original EM images, 20 single-object GT images and 20 multi-object GT images. EMDS-5 can realize to evaluate image preprocessing, image segmentation, feature extraction, image classification and image retrieval functions. In order to prove the effectiveness of EMDS-5, for each function, we select the most representative algorithms and price indicators for testing and evaluation. The image preprocessing functions contain two parts: image denoising and image edge detection. Image denoising uses nine kinds of filters to denoise 13 kinds of noises, respectively. In the aspect of edge detection, six edge detection operators are used to detect the edges of the images, and two evaluation indicators, peak-signal to noise ratio and mean structural similarity, are used for evaluation. Image segmentation includes single-object image segmentation and multi-object image segmentation. Six methods are used for single-object image segmentation, while k-means and U-net are used for multi-object segmentation. We extract nine features from the images in EMDS-5 and use the Support Vector Machine (SVM) classifier for testing. In terms of image classification, we select the VGG16 feature to test SVM, k-Nearest Neighbors, Random Forests. We test two types of retrieval approaches: texture feature retrieval and deep learning feature retrieval. We select the last layer of features of VGG16 network and ResNet50 network as feature vectors. We use mean average precision as the evaluation index for retrieval. EMDS-5 is available at the URL:https://github.com/NEUZihan/EMDS-5.git.

FUSION OF MULTIMODAL BIOMETRICS OF FINGERPRINT, IRIS AND HAND WRITTEN SIGNATURES TRAITS USING DEEP LEARNING TECHNIQUE

Due to expanding interest for the data security and safety guidelines everywhere, biometricauthentication technology has been generally utilized in our regular day to day existence. With respects to this,multi-modal biometric innovation has acquired attention and became famous because of the capacity to aovercome the drawbacks of uni-model biometric frameworks. In Present research, novel multi biometricsrecognition proof solution is developed, that depends to deep learning techniques for perceiving human utilizingmulti biometric traits of Iris pattern, finger print data and offline signature biometrics. Framework of designdepends on Deep Neural Networks (DNNs), for separating the parameters &amp; classification of the image utilizingsoft max based technique. To foster the framework, deep learning models are joined iris, finger print and off-linesignature. To construct the VGG-19 network was utilized, and Adam streamlining technique has been applied forunmitigated to measure the degree of inequality was utilized as a misfortune work. A few strategies to stay awayfrom overfitting were applied, like picture increase and drop-out procedures. For combining the deep learningnetworks, different combinations are utilized to investigate the impact of techniques on acknowledgmentexecution, accordingly component and score-level combination approach was applied. The exhibition of proposedframework is experimentally by directing a few trials to the SDUMLA-HMT data set, which is multi-modalbiometric data set. Acquired outcomes showed that involving triple biometrics attributes in biometric distinguishedproof frameworks got preferred outcomes over utilizing a couple biometric characteristics. The outcomesadditionally shows that our methodology serenely beat other condition of- - the-craftsmanship techniques byaccomplishing a precision of 99.11% on an element degree combination procedures and of 99.21 percentaccuracy of various strategy for fusion at score level.

Accuracy improving of pre-trained neural networks by fine tuning

Methods of accuracy improving of pre-trained networks are discussed. Images of ships are input data for the networks. Networks are built and trained using Keras and TensorFlow machine learning libraries. Fine tuning of previously trained convoluted artificial neural networks for pattern recognition tasks is described. Fine tuning of VGG16 and VGG19 networks is done by using Keras Applications. The accuracy of VGG16 network with fine tuning of the last convolution unit increased from 94.38% to 95.21%. An increase is only 0.83%. The accuracy of VGG19 network with fine tuning of the last convolution unit increased from 92.97% to 96.39%, which is a 3.42% increase.

Recognition of abnormal body surface characteristics of oplegnathus punctatus

To identify the abnormal characteristics of the oplegnathus punctatus is great importance to the detection of iridovirus disease in the breeding environment. In this paper, an advanced neural network model to identify the characteristics of the oplegnathus punctatus and predict its different periods of suffering from iridovirus disease is proposed based on the establishment of a data set. First of all, a standard format data set of oplegnathus punctatus and an abnormal format date set are established in order to verify the effectiveness of the method in this paper. And then, the feature extraction fusion method is used for preprocessing in terms of the abnormal format data set, which combines the edge features extracted by the improved multi-template Sobel operator and the color features extracted by the HSV model. Finally, an improved VGG-GoogleNet network recognition model comes into being through the fusion and improvement of the VGG and GoogleNet neural network structure. The experiments results show that the prediction accuracy rate for oplegnathus punctatus suffering from iridovirus disease in the the abnormal format data set and the standard format data set are improved, which reach 98.55% and 69.18%.

Detecting potholes in asphalt pavement under small-sample conditions based on improved faster region-based convolution neural networks

Pavement potholes have low detection accuracy under the condition of small samples. To address this issue, we propose a method for efficient and accurate pothole detection under small-sample conditions based on an improved Faster R-CNN (Region-based Convolution Neural Networks). First, images consisting of different pothole shapes and sizes were acquired from different sources and then augmented and denoised to obtain an image set. Second, two representative target-detection models, Faster R-CNN and YOLOv3, were tested. The detection results indicate that Faster R-CNN achieves better detection performance. Furthermore, to overcome inconsistencies (missed detections and inaccurate position estimations), the feature extraction layers of VGG16, ZFNet, and ResNet50 networks were used in combination with Faster R-CNN. The results showed that the VGG16+Faster R-CNN fusion model yielded superior accuracy. Finally, the detection accuracy improved to 0.8997 after adjusting the size of the candidate frame, which also enabled successful detection of previously missed targets.

Deep feature loss to denoise OCT images using deep neural networks.

.Significance: Speckle noise is an inherent limitation of optical coherence tomography (OCT) images that makes clinical interpretation challenging. The recent emergence of deep learning could offer a reliable method to reduce noise in OCT images.Aim: We sought to investigate the use of deep features (VGG) to limit the effect of blurriness and increase perceptual sharpness and to evaluate its impact on the performance of OCT image denoising (DnCNN).Approach: Fifty-one macula-centered OCT pairs were used in training of the network. Another set of 20 OCT pair was used for testing. The DnCNN model was cascaded with a VGG network that acted as a perceptual loss function instead of the traditional losses of and . The VGG network remains fixed during the training process. We focused on the individual layers of the VGG-16 network to decipher the contribution of each distinctive layer as a loss function to produce denoised OCT images that were perceptually sharp and that preserved the faint features (retinal layer boundaries) essential for interpretation. The peak signal-to-noise ratio (PSNR), edge-preserving index, and no-reference image sharpness/blurriness [perceptual sharpness index (PSI), just noticeable blur (JNB), and spectral and spatial sharpness measure (S3)] metrics were used to compare deep feature losses with the traditional losses.Results: The deep feature loss produced images with high perceptual sharpness measures at the cost of less smoothness (PSNR) in OCT images. The deep feature loss outperformed the traditional losses ( and ) for all of the evaluation metrics except for PSNR. The PSI, S3, and JNB estimates of deep feature loss performance were 0.31, 0.30, and 16.53, respectively. For L1 and L2 losses performance, the PSI, , and JNB were 0.21 and 0.21, 0.17 and 0.16, and 14.46 and 14.34, respectively.Conclusions: We demonstrate the potential of deep feature loss in denoising OCT images. Our preliminary findings suggest research directions for further investigation.

High-Level Descriptors for Fall Event Detection Supported by a Multi-Stream Network

The need for assertive video classification has been increasingly in demand. Especially for detecting endangering situations, it is crucial to have a quick response to avoid triggering more serious problems. During this work, we target video classification concerning falls. Our study focuses on the use of high-level descriptors able to correctly characterize the event. These descriptor results will serve as inputs to a multi-stream architecture of VGG-16 networks. Therefore, our proposal is based on the analysis of the best combination of high-level extracted features for the binary classification of videos. This approach was tested on three known datasets, and has proven to yield similar results as other more consuming methods found in the literature.

Entropy Slicing Extraction and Transfer Learning Classification for Early Diagnosis of Alzheimer Diseases with sMRI

Alzheimer’s Disease (AD) is an irreversible neurogenerative disorder that undergoes progressive decline in memory and cognitive function and is characterized by structural brain Magnetic Resonance Images (sMRI). In recent years, sMRI data has played a vital role in the evaluation of brain anatomical changes, leading to early detection of AD through deep networks. The existing AD problems such as preprocessing complexity and unreliability are major concerns at present. To overcome these, a model ( FE ES C TL ) has been proposed with an entropy slicing for feature extraction and Transfer Learning for classification. In the present study, the entropy image slicing method is attempted for selecting the most informative MRI slices during training stages. The ADNI dataset is trained on Transfer Learning adopted by VGG-16 network for classifying the AD with normal individuals. The experimental results reveal that the proposed model has achieved an accuracy level of 93.05%, 86.39%, 92.00% for binary classifications (AD/MCI, MCI/CN, AD/CN) and 93.12% for ternary classification (AD/MCI/CN), respectively, and henceforth the efficiency in diagnosing AD is proved through comparative analysis.

Application of deep learning for semantic segmentation of sandstone thin sections

Sedimentary petrology is the basis for most mineral and textural identification in sandstones. Automating mineralogical interpretation of an entire thin section image has many practical applications, including improved geological understanding, input of spatial distribution of mineralogy and grain size for petrophysical evaluations, and integration with 3D imaging modalities (micro-CT, nano-CT). We investigate the application of Convolutional Neural Network (CNN) based supervised semantic segmentation methods for predicting pixel-scale mineralogy using 2D RGB images of sandstones acquired by transmission light microscopy. Models were trained to interpret a simple binary pore-mineral (grain) segmentation and a 10-class segmentation (porosity, quartz, feldspar, rock fragments, carbonate grains, opaque grains, quartz cement, carbonate cement, clay cement, and hydrocarbons filling pores). For the 2-class classification framework to distinguish between pores and minerals, most models lead to satisfactory results with acceptable accuracy. For the 10-class classification framework, models trained with Deeplab V3+ Resnet-18 network yield more continuous results compared to those based on VGG networks. We conclude that the effectiveness of the models, in predicting a petrology class in a thin section, strongly correlates with the amount of labeled data available to train the model to interpret the class in question. Semantic segmentation models, based on CNNs, can produce encouraging results for a 10-class petrological classification framework of an entire thin section image and thus provide a complete scene understanding which is difficult to produce manually.