A hyperspectral anomaly detection framework based on segmentation and convolutional neural network algorithms

The construction of statistical shape models (SSMs) is an important method in the field of medical image segmentation. Most SSMs are constructed by using traditional modeling methods based on principal component analysis (PCA), which cannot fully present the true deformation ability of models. To solve the insufficient deformation ability of SSMs, we propose a stacked autoencoder (SAE) neural network to construct a multi-resolution multi-organ shape model based on mouse micro-CT images, which can express more linear and non-linear deformations than SSMs based on PCA. The main advantage of this method is that the SAE neural network is simple and flexible and it can learn more deformation modes from training data. We have quantitatively compared the modeling performance of this method with the constructed SSMs based on PCA in terms of model generalization and specificity.

Hybrid models based on genetic algorithm and deep learning algorithms for nutritional Anemia disease classification

The rapid development of Internet technology has led to increasingly complex web service systems. The resulting large number of component interactions pose a challenge to anomaly detection, which is realized primarily by operation and maintenance (OM) personnel through the deployment, management, and monitoring of various key performance indicators (KPIs). Anomalous behaviors during daily OM often cause problems in KPI data; these problems include high noise, high dimensionality, and large-scale data streams. In addition, anomalies in KPI data occur infrequently and are of various types. These factors are the cause of the very low accuracy currently observed in conventional machine learning methods for detecting anomalies in large OM systems. Hence, a convolutional long short-term memory (C-LSTM) neural network is presented in this study to detect anomalies in small datasets that contain a variety of anomalies. First, a sliding window is used to preprocess the KPI data. Then, a C-LSTM neural network, which combines the features of the convolutional neural network (CNN) and LSTM algorithms, is employed to effectively model the time and numerical information contained in the preprocessed KPI data. Finally, the C-LSTM algorithm is tested on the datasets used in the competition of the Artificial Intelligence for Information Technology Operations (AIOPs) Active Network Management (ANM) 2018 Fall Project. The results show that the C-LSTM prediction algorithm outperforms the conventional LSTM and CNN algorithms in terms of its capacity to detect anomalies in small datasets that contain various anomalies, with a 12.90% higher accuracy, 5.68% higher recall, and 9.58% higher F1-score.

Hyperspectral imagery has high spectral resolution, and spectrum of it has always been non-linear. The traditional classification methods cannot get better result when the number of samples is small. Combined with the theory of deep learning, a new semi-supervised method based on stacked autoencoders (SAE) is proposed for hyperspectral imagery classification. Firstly, with stacked autoencoders, a deep network model is constructed. Then, unsupervised pre-training is carried combined SOFTMAX classifier with unlabeled samples. Finally, fine-tuning the network model with small labeled samples, the SAE-based classifier can be got to learn implicit feature of spectrum of hyperspectral imagery and achieve classification of hyperspectral imagery. According to comparative experiments, the results indicate that the proposed method is effective to improve the hyperspectral imagery classification accuracy in case of small samples.

To develop and test a deep learning approach named Convolutional Neural Network (CNN) for automated screening of T2 -weighted (T2 WI) liver acquisitions for nondiagnostic images, and compare this automated approach to evaluation by two radiologists. We evaluated 522 liver magnetic resonance imaging (MRI) exams performed at 1.5T and 3T at our institution between November 2014 and May 2016 for CNN training and validation. The CNN consisted of an input layer, convolutional layer, fully connected layer, and output layer. 351 T2 WI were anonymized for training. Each case was annotated with a label of being diagnostic or nondiagnostic for detecting lesions and assessing liver morphology. Another independently collected 171 cases were sequestered for a blind test. These 171 T2 WI were assessed independently by two radiologists and annotated as being diagnostic or nondiagnostic. These 171 T2 WI were presented to the CNN algorithm and image quality (IQ) output of the algorithm was compared to that of two radiologists. There was concordance in IQ label between Reader 1 and CNN in 79% of cases and between Reader 2 and CNN in 73%. The sensitivity and the specificity of the CNN algorithm in identifying nondiagnostic IQ was 67% and 81% with respect to Reader 1 and 47% and 80% with respect to Reader 2. The negative predictive value of the algorithm for identifying nondiagnostic IQ was 94% and 86% (relative to Readers 1 and 2). We demonstrate a CNN algorithm that yields a high negative predictive value when screening for nondiagnostic T2 WI of the liver. 2 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2018;47:723-728.

In this paper, we designed an end-to-end spectral–spatial residual network (SSRN) that takes raw 3-D cubes as input data without feature engineering for hyperspectral image classification. In this network, the spectral and spatial residual blocks consecutively learn discriminative features from abundant spectral signatures and spatial contexts in hyperspectral imagery (HSI). The proposed SSRN is a supervised deep learning framework that alleviates the declining-accuracy phenomenon of other deep learning models. Specifically, the residual blocks connect every other 3-D convolutional layer through identity mapping, which facilitates the backpropagation of gradients. Furthermore, we impose batch normalization on every convolutional layer to regularize the learning process and improve the classification performance of trained models. Quantitative and qualitative results demonstrate that the SSRN achieved the state-of-the-art HSI classification accuracy in agricultural, rural–urban, and urban data sets: Indian Pines, Kennedy Space Center, and University of Pavia.

Various existing anomaly detection (AD) technologies focus on the background feature extraction and suppression, which serves as a crucial step to extrude anomalies from the hyperspectral imagery (HSI). In this article, motivated by the advantages of the joint sparse representation (JSR) model for adaptive background base selection, a robust background feature extraction method through homogeneous region-based JSR is proposed and used for AD. By segmenting the scene from the spatial domain through an eight-connected region division operation based on the clustering result, a series of nonoverlapping homogeneous regions each sharing a common sparsity pattern are obtained. After discarding small regions, JSR is performed on each region with the dictionary constituted by the overall spectral items in the corresponding cluster. By calculating the usage frequency of dictionary atoms, the most representative background bases describing each background cluster are adaptively selected and then combined into the global background bases. In addition, considering the interference of noise on detection accuracy, an energy deviation-based noise estimation strategy is presented by analyzing the residual obtained from JSR. Finally, the anomaly response of each pixel is measured by comparing its projection energy obtained from the background orthogonal subspace projection with the noise energy in its corresponding region. The proposed method overcomes the shortcomings of traditional neighborhood-based JSR in the common sparsity pattern and anomaly proportion. The spatial characteristics of HSI are fully explored. Furthermore, the interference of noise on detection accuracy is eliminated. Experiments on four HSI data sets demonstrate the superiority of the proposed method.

Identification of crops is an important topic in the agricultural domain. Hyperspectral remote sensing data are very useful for crop feature extraction and classification. Remote sensing data is an unstructured data and Convolutional Neural Network (CNN) can work on unstructured data efficiently. This paper presents an evaluation of CNN for crop classification using the Indian Pines standard dataset obtained from the AVIRIS sensor and the study area dataset obtained from the EO-1hyperion sensor. Optimized CNN has been tuned by training the model on different parameters. It has been compared with two classification algorithms: Deep Neural Network (DNN) and Convolutional Autoencoder. According to the test results, the proposed optimized CNN model provided better results as compared to the other two methods. CNN has given 97 ± 0.58% overall accuracy for the Indian Pines standard dataset and 78 ± 2.43% for our study area dataset.

ABSTRACTAnomaly detection is an active research topic in hyperspectral remote sensing and has been applied in many areas including environmental monitoring, urban survey, mineral mapping, and national security. Usually, it makes a detection decision without any prior target information or background information. Several anomaly detection algorithms (e.g. Reed–Xiaoli detector, blocked adaptive computationally efficient outlier nominator and random-selection-based anomaly detector) have been developed, which rely on estimating background information only from a hyperspectral image without considering target information in making a detection decision. These methods may be efficient in general but sometimes with high false alarm rate (FAR). In order to reduce FAR, this study proposes a novel method that incorporates both background and target information, derived from the hyperspectral imagery, into anomaly detection algorithms. The target information is helpful to detect anomalies as outliers. With a scene of real airborne visible infrared imaging spectrometer data, the experimental results demonstrate that the proposed method has produced better detection results and higher time efficiency compared with those using the traditional algorithms that only consider background information.

In order to overcome the problem of inaccuracy in feature extraction when the corner features are not obvious in traditional corner detection methods, a face feature extraction method based on deep learning convolutional neural network is proposed in this paper. The key point of this method is the construction algorithm of convolution network with single image as input data. In this paper, the convolutional neural network implemented by Python code includes three convolution layers, three lower sampling layers and one full connection layer. The filter structure is designed and the characteristic image is obtained. The convolution layer uses the Reul function as the activation function, while the lower sampling layer uses the maximum pooling and ruul activation functions. The experimental results show that the method based on the different levels of image features extracted by deep learning network as the basis for image matching, and can still achieve good recognition effect when the traditional features such as corners are not clear.

Hyperspectral anomaly detection based on anomalous component extraction framework

Batik is an Indonesian cultural heritage that should be preserved. Over time, many batik motifs have sprung up, which can lead to mutual claims between craftsmen. Therefore, it is necessary to create a system to measure the similarity of a batik motif. This research is focused on making Content-Based Image Retrieval (CBIR) on batik images. The dataset used in this research is big data Batik images. The authors used transfer learning on several pre-trained models and used Convolutional Neural Network (CNN) Autoencoder from previous studies to extract features on all images in the database. The extracted features calculate the Euclidean distance between the query and all images in the database to retrieve images. The image closest to the query will be retrieved according to the number of r, namely 3, 5, 10, or 15. Before the image is retrieved, the retrieval system is used to re-ranked with K-Nearest Neighbor (KNN), which classifies the retrieved image. The results of this study prove that MobileNetV2 + KNN is the best model in terms of Image Retrieval Batik, followed by InceptionV3 and VGG19 as the second and third ranks. Moreover, CNN Autoencoder from previous research and InceptionResNetV2 are ranked fourth and fifth. In this study, it was also found that the use of KNN re-ranking can increase the precision value by 0.00272. For further research, deploying these models, especially for MobileNetV2 is an approach for seeing a major impact on batik craftsmanship for decreasing batik motif plagiarism.

The general objective of anomaly detection (AD) in hyperspectral imagery is to detect full-pixel targets. To meet this purpose, the global AD methods can achieve more reliable results than the local methods in terms of time and accuracy. The kernel-based Support Vector Data Description (SVDD) has recently received great attention in the hyperspectral AD applications. This paper presents a global SVDD-based method for autonomous full-pixel AD. The method consists of three steps: clustering, background modeling, and autonomous AD. Experimental results on a hyperspectral dataset show the superiority of the proposed method comparing to the global-based SVDD method.

Due to the data-hungry nature of deep neural networks and the high dimensionality of hyperspectral imagery (HSI) data, the scarcity of training samples remains a significant challenge for HSI classification based on deep learning. Recently, the pairing or recombining of samples has proven to be an efficient approach to increase the amount of input training data, and accordingly, the architecture of a network trained by these samples has to be redesigned, since the input is no longer a single sample. Following this strategy, in this article, we propose a spatial–spectral relation network (SS-RN) for HSI classification with limited training samples. The SS-RN takes advantage of the relation network architecture to precisely capture the profound similarity between samples. Additionally, to make SS-RN more suitable for HSI classification, an entire three-dimensional (3-D) neighborhood instead of the isolated pixel is considered to explore both spectral and spatial information thoroughly. Multiple support samples are also selected for each class to make sure that the extracted features are stable enough to avoid the intraclass similarity and interclass dissimilarity problem of HSI. The network is composed on the basis of 3-D convolutional neural network blocks with the aim of extracting spatial–spectral features. Extensive experiments on three widely used HSI datasets demonstrate that the proposed method can achieve better classification accuracy than conventional deep learning methods with limited labeled training samples.

Due to the limitations of hyperspectral imaging systems, hyperspectral imagery (HSI) often suffers from poor spatial resolution, thus hampering many applications of the imagery. Hyperspectral super-resolution refers to fusing HSI and MSI to generate an image with both high spatial and high spectral resolutions. Recently, several new methods have been proposed to solve this fusion problem, and most of these methods assume that the prior information of the Point Spread Function (PSF) and Spectral Response Function (SRF) are known. However, in practice, this information is often limited or unavailable. In this work, an unsupervised deep learning-based fusion method - HyCoNet - that can solve the problems in HSI-MSI fusion without the prior PSF and SRF information is proposed. HyCoNet consists of three coupled autoencoder nets in which the HSI and MSI are unmixed into endmembers and abundances based on the linear unmixing model. Two special convolutional layers are designed to act as a bridge that coordinates with the three autoencoder nets, and the PSF and SRF parameters are learned adaptively in the two convolution layers during the training process. Furthermore, driven by the joint loss function, the proposed method is straightforward and easily implemented in an end-to-end training manner. The experiments performed in the study demonstrate that the proposed method performs well and produces robust results for different datasets and arbitrary PSFs and SRFs.