Generally, two types of approaches, namely, data leveling [18–20] and algorithm lev- elling [9, 21, 22] are employed to address the imbalanced datasets problem. Over- or down-sampling methods used at the data level attempt to balance the majority and minority class proportions by data resampling to address the imbalanced problem. How- ever, this approach can easily lead to redundant or missing information and thus affect the classification performance [20, 21, 23]. By contrast, the cost-sensitive approach using algorithm leveling has a distinct advantage because it makes full use of the original data [9, 21, 22]. Meanwhile, deep convolutionalneuralnetwork (CNN) models have demon- strated extraordinary performance in medical image recognition tasks [24–29]. In this study, we combine a representative deep learning CNN (deep residual network [30]) and a cost-sensitive data-balancing method to present an effective cost-sensitive residual CNN (CS-ResCNN) for the ophthalmic imbalanced dataset problem. By using a grid- search analysis procedure, we demonstrate the robustness and effectiveness of the CS- ResCNN. Finally, we develop and deploy a web-based computer-aided diagnosis (CAD) software based on our proposed method for patients and ophthalmologists in clinical application.
We present a new deep learning architecture Bi-CNN-MI for paraphrase identification (PI). Based on the insight that PI requires compar- ing two sentences on multiple levels of granu- larity, we learn multigranular sentence repre- sentations using convolutionalneuralnetwork (CNN) and model interaction features at each level. These features are then the input to a logistic classifier for PI. All parameters of the model (for embeddings, convolution and clas- sification) are directly optimized for PI. To ad- dress the lack of training data, we pretrain the network in a novel way using a language mod- eling task. Results on the MSRP corpus sur- pass that of previous NN competitors.
Transportation problem is one of the crucial issues faced by many big cities.With the development of computer technology and artificial intelligence technology, intelligent transportation will become a hot research topic. Vehicle recognition is one of the key technologies of intelligent transportation.Intelligent vehicle identification in vehicle management, illegal escape vehicle, a vehicle inspection surveillance and many other issues have played a key role. At present domestic and international mainstream vehicle recognition methods are as follows: sense coil detection, infrared detection method, the dynamic piezoelectric detection method, etc. Although the recognition accuracy is still relatively high, the installation process will affect the traffic order normal operation, equipment is also easily to damage, and difficult and expensive maintenance.Vehicle recognition method based on image processing acquires the vehicle image through high precision industrial camera and image acquisition card. Then computer is used to simulate the function of human visual effect, analyze to extract the required information, such as vehicle license, color, shape and other features. Finally, we use pattern recognition method to distinguish between different models. In this paper, the method of vehicle recognition is based on image processing method, using convolutionalneuralnetwork (CNN) algorithm, which is trained by a large number of sample models, thus the accuracyis improved.This method is essentially different from the traditionalalgorithm in the aspect of recognition rate and with a high quality of the recognition ratein a variety of scenarios.
As shown in Table 2, our proposed model has significant improvements in all validation datasets. The lowest improvement is nearly 1% of accuracy and the highest improvement is over 6% of accuracy. These improve- ments are quite high in comparison with other approaches such as finding good representations for sequences or feature selection which were applied before. It showed that features extracted by convolutional layers of the convolutionalneuralnetwork are very useful for the classifier to classify sequences into true categories.
In this paper we proposes novel algorithm to detect station logo, which is based on deep learning convolutionalneuralnetwork and target detection tool SSD. Firstly, construct an experimental environment of deep learning by optimizing the algorithm parameters. Secondly, the algorithm is modeled by collecting diversified station caption data. Finally, the model is applied to the real application environment after reaching the predetermined detection accuracy.
time series data and promises a high accuracy output. The social media Facebook Research division head and Father of network Architecture YannLeCun uses a new architecture which is good at object recognition in image dataset called the ConvolutionalNeuralNetwork (CNN). The convolutional technique is showing a great success in image processing like multilayer perceptron feed forward neural networks. Also this technique is capable of scaling with data and model size and the model could be well trained with back propagation algorithm. This fundamental idea and the requirement leads to the significance of deep learning as the development of CNNs with large number of layers, which has promised to produce a high accurate detection and classification rate on image and video content. Deep learning methods have several well-defined computational models, which consist of multiple computational layers to learn representations of input data with multiple abstraction levels. Deep learning is a well known classification and prediction technique which is used for constructing and training neuralnetwork models that are considered highly promising best models for image promising. Deep learning techniques have the word deep in the sense that its input data should be passed and processed through a series of nonlinear transformations before it comes out of the output layer. For extracting the features from the images, deep learning networks employ an automated technique. The trained data follows the well defined training procedure and it identifies the new and required pattern from the image given as input to network. Image recognition is another interesting area of application. The images are represented as a 2D array of pixels. Each pixel with RGB channels or gray scale is feed directly into a convolutionalneuralnetwork which is trained end-to-end. A CNN consists of alternating layers of convolutions called convolutional layers, and also it contains pooling and sub sampling layers. The result of this is a deep abstract representation of images at the output layer thus CNNs or convents is thus it a powerful tool for classifying the contents of images. Therefore the deep network leads to the success of:
At present times, biometric recognition, like face recog- nition, is being used as the main security barrier in various projects, making the need for it to be as fast and accurate as possible obvious. The state of the art methods are using a trained convolutionalneuralnetwork (CNN) on a huge data set in order to make it efficient and precise. The problem introduced, is that the output of the training is not something that can be analyzed and its workings understood, but has to be taken as always true. As this is a really important piece in the security process, understanding how it actually works is crucial. In this paper a step towards that will be attempted.
Abstract- ConvolutionalNeuralNetwork (CNN) is transforming the field of medical diagnosis. CNN can help doctors make faster, more accurate diagnosis by providing automatic learning techniques for predicting the common patterns from the medical image data. Human expert provides limited interpretation of medical images due to its subjectivity, complexity and extensive variations across the image. CNN is able to provide state of the art solutions with good accuracy for medical imaging and is powered by the increasing availability of healthcare data. Major disease areas that use CNN includes cancer, dermatology, neurology and cardiology. This paper focuses on the use of different CNN architectures based on their performance for accurate medical diagnosis. We also discuss the current status of CNN applications in healthcare and its various limitations.
Abstract:- Detection of rice pest and diseases, and proper management and control of pest infested rice fields may result to a higher rice crop production. Using modern technologies, like smart phones, farmers can be aided in detecting and identifying the type of pests and diseases found in their rice fields. ConvolutionalNeuralNetwork using r language are used to find the diseases in rice by using images of disease leaves. The disease images are collected from the UCI Machine Learning Repository contain the three types of diseases namely Bacterial Leaf Blight, Brown Spot, Leaf Smut.
Convolutionalneuralnetwork (CNN) is an essential model to achieve high accuracy in various machine learning applications, such as image recognition and natural language processing. One of the important issues for CNN acce- leration with high energy efficiency and processing performance is efficient data reuse by exploiting the inherent data locality. In this paper, we propose a novel CGRA (Coarse Grained Reconfigurable Array) architecture with time- domain multithreading for exploiting input data locality. The multithreading on each processing element enables the input data reusing through multiple computation periods. This paper presents the accelerator design performance analysis of the proposed architecture. We examine the structure of memory subsystems, as well as the architecture of the computing array, to supply re- quired data with minimal performance overhead. We explore efficient archi- tecture design alternatives based on the characteristics of modern CNN con- figurations. The evaluation results show that the available bandwidth of the external memory can be utilized efficiently when the output plane is wider (in earlier layers of many CNNs) while the input data locality can be utilized maximally when the number of output channel is larger (in later layers).
Abstract. Technological development in recent years has generated the constant need to digitalize and analyze data, where handwritten digit recognition is a popular problem. This paper focuses on the creation of two handwritten digit datasets and their use to train a ConvolutionalNeuralNetwork (CNN) to classify them, also, a proposed extra preprocessing technique is applied to the images of one of the data sets. Experiments show that the proposed preprocessing technique lead to obtain accuracies above 98%, which were higher than the values obtained with the dataset without the additional preprocessing.
to evaluate and improve the accuracy and efficiency of the model. Jie Chang Et al. [18] have done segmentation of MR images of the brain using CNN for developing a method of automatic brain tumor detection. They have developed a two- way path model which contains one average pooling layer and other max pooling layer in different paths. Then, finally the CNN model is combined to a fully connected layer to predict optimized results. As MRI scans are very useful in detecting inner body dysfunction, the need of automatic region detection becomes an essential in the medical field. Pre-processing-Not having properly normalized and quantifiable pixel intensities interpretation is one of very big defects in MRI data. So, we need to preprocess the training data in order to extract meaningful quantifiable data. With the growing number of microscopic images automatic Nano particle detection has become an essential task to achieve. Ayse Betul Oktay Et al. [19] have discussed a method to detect Nano particle in microscopic images using segmentation. They have proposed a method for the detection of Nano-particles and detection of their shapes and sizes with the help of deep learning algorithms. This method employees multiple output CNN and has two outputs. First is the detection output which gives the location of the Nano particle in the input image. Other is the segmentation output which outputs the boundary of segmented Nano particles. The final sizes of the Nano particles are determined by the Hough algorithm that works on the segmentation outputs. Here they have used MOCNN i.e. multiple output Convolutionalneuralnetwork for the purpose of detecting, localizing and segmenting of Nano particles found in microscopic images [19]. MOCNN takes an input image as a window and it produces two outputs for it, from which one output tells us the location of the Nano particle in the image and second output tells us the distance of the object boundary to the window center. Despite very accurate and one of the best ways of image segmentation by CNN, it does have some uncertainties which nobody likes to tackle. Guotai Wang Et al. [20] have analyzed these different kinds of uncertainties for CNN based 2D and 3D image segmentation. Additionally, they have analyzed a test time augmentation-based uncertainty to analyze the effect of different kind of image transformations on
Road is an important kind of basic geographic information. Road information extraction plays an important role in traffic management, urban planning, automatic vehicle navigation, and emergency management. With the development of remote sensing technology, the quality of high-resolution satellite images is improved and more easily obtained, which makes it possible to use remote sensing images to locate roads accurately. Therefore, it is an urgent problem to extract road information from remote sensing images. To solve this problem, a road extraction method based on convolutionalneuralnetwork is proposed in this paper. Firstly, convolutionalneuralnetwork is used to classify the high-resolution remote sensing images into two classes, which can distinguish the road from the non-road and extract the road information initially. Secondly, the convolutionalneuralnetwork is optimized and improved from the training algorithm. Finally, because of the influence of natural scene factors such as house and tree shadow, the non-road noise still exists in the road results extracted by the optimized convolutionalneuralnetwork method. Therefore, this paper uses wavelet packet method to filter these non-road noises, so as to accurately present the road information in remote sensing images. The simulation results show that the road information of remote sensing image can be preliminarily distinguished by convolutionalneuralnetwork; the road information can be distinguished effectively by optimizing convolutionalneuralnetwork; and the wavelet packet method can effectively remove noise interference. Therefore, the proposed road extraction method based on convolutionalneuralnetwork has good road information extraction effect.
Of the other sentence models, the NBoW is a shallow model and the RNN has a linear chain structure. The subgraphs induced in the Max- TDNN model have a single fixed-range feature ob- tained through max pooling. The recursive neuralnetwork follows the structure of an external parse tree. Features of variable range are computed at each node of the tree combining one or more of the children of the tree. Unlike in a DCNN, where one learns a clear hierarchy of feature orders, in a RecNN low order features like those of sin- gle words can be directly combined with higher order features computed from entire clauses. A DCNN generalises many of the structural aspects of a RecNN. The feature extraction function as stated in Eq. 6 has a more general form than that in a RecNN, where the value of m is generally 2. Likewise, the induced graph structure in a DCNN is more general than a parse tree in that it is not limited to syntactically dictated phrases; the graph structure can capture short or long-range seman- tic relations between words that do not necessar- ily correspond to the syntactic relations in a parse tree. The DCNN has internal input-dependent structure and does not rely on externally provided parse trees, which makes the DCNN directly ap- plicable to hard-to-parse sentences such as tweets and to sentences from any language.
Object Recognition is an important part of modern intelligent machines and systems. It is used in many applica- tions related to multiple fields such as character recognition for mail sorting service [1], traffic monitoring [2], surveillance for security purposes [3], self-driving vehicle [4], human behaviour analysis [5], and medical imag- ing [6]. Object recognition has been performed generally using volumetric parts (i.e. generalized cylinders, geons and super-quadrics) [7], appearance based (i.e. edges, lines, corners) [8], pattern recognition (i.e. SIFT, HOG) [9], graph-based [10], and learning based methods [11]. Although the techniques mentioned above are still used in many applications, after the success of the award winning deep learning architecture alexnet [12], ConvolutionalNeural Networks (CNN) become ubiquitous. In recent years, deep learning has achieved out- standing results in 2D object recognition [13]. These results motivate researchers to apply deep learning me- thods for 3D object recognition, and some of them provide better results in comparison to state-of-art (SoA) ap- proaches.
powerful for unbounded dependencies, but tweets are short; the sentiment of a tweet is usually de- termined by one part of it and unlike RNN/LSTM, convolution plus max pooling can learn to focus on that. Recursive architectures like the Recursive Neural Tensor Network (Socher et al., 2013). as- sume some kind of hierarchical sentence structure. This structure does not exist or is hard to recognize for many noisy tweets.
The present work aims to address this ever increasing gap between the volumes of actual data generated and the volume that can be reasonably inspected manually. It is laborious and time consuming to scrutinize the salient events from the large video databases. We introduce smart surveillance by using video summarization for various applications. Thereafter analyzed via ConvolutionalNeural Networks (CNN). Using this analysis, features are extracted and calculated, which are used for generation of summarized videos.
Abstract: Looking at the population increase in India, waste generation and segregation is a major issue in the current scenario. Tonnes of mixed waste is dumped without segregating it properly which leads to problem in decomposition. Due to this mixed waste several other problems arise over a period of time. To avoid this, waste segregation at least at the basic level is very much needed. We have implemented a system based on ConvolutionalNeural Networks. The basic idea is that when the waste is to be dumped in the garbage bin, the system will identify the type of waste and will open the dustbin of that category accordingly. Using this system, it becomes easier the segregate waste at the basic level. We have four categories in which waste will be segregated namely, glass, paper, plastic, metal. Four distinct dustbins along with servo motors will be used for the same.
During training, the only pre-processing step is to get the mean value of RGB, which is on the computed training data. After that the images are moved to the stack of convolutional layers. Three fully connected layers follow the stack of convolutional layers. The FeatureDetector interface is used to find interest points. The SurfFeatureDetector and its function will perform the detection process. The function drawKeypoints is used to draw the detected keypoints. The DescriptorExtractor is an interface used to find the feature vector correspondent to the keypoints. Use SurfDescriptorExtractor and its function computes the performance of the required calculations. Use a BFMatcher to match the features vector. The function drawMatches is used to draw the detected matches. The CascadeClassifier class is used to detect objects in the video stream. Initially we have to load the.xml classifier file which contains the images that are to be loaded. DetectMultiScale function is used to perform detection.