Object Classification Accuracy Research Articles

Structural Image De-Identification for Privacy-Preserving Deep Learning

Due to the risk of data leakage while training deep learning models in a shared environment, we propose a new privacy-preserving deep learning (PPDL) method using a structural image de-identification approach for object classification. The proposed structural image de-identification approach is designed based on the fact that the degree of structural distortion of an image object has the greatest impact on human’s perceptual system. Thus, by modifying only the structural parts of the original one using order preserving encryption(OPE), the proposed structural image de-identification approach decreases only the recognition rate by human. From the experimental results using different standard datasets, we show that the object classification accuracy of the proposed structural image de-identification method is almost the same as the deep learning performance for non-encrypted images, without revealing the original image contents including sensitive information. Also, by handling the trade-off between object classification accuracy and privacy protection for the de-identified image, we experimentally find the optimal size of input image for the proposed structural image de-identification approach.

Underwater Object Classification in Sidescan Sonar Images Using Deep Transfer Learning and Semisynthetic Training Data

Sidescan sonars are increasingly used in underwater search and rescue for drowning victims, wrecks and airplanes. Automatic object classification or detection methods can help a lot in case of long searches, where sonar operators may feel exhausted and therefore miss the possible object. However, most of the existing underwater object detection methods for sidescan sonar images are aimed at detecting mine-like objects, ignoring the classification of civilian objects, mainly due to lack of dataset. So, in this study, we focus on the multi-class classification of drowning victim, wreck, airplane, mine and seafloor in sonar images. Firstly, through a long-term accumulation, we built a real sidescan sonar image dataset named SeabedObjects-KLSG, which currently contains 385 wreck, 36 drowning victim, 62 airplane, 129 mine and 578 seafloor images. Secondly, considering the real dataset is imbalanced, we proposed a semisynthetic data generation method for producing sonar images of airplanes and drowning victims, which uses optical images as input, and combines image segmentation with intensity distribution simulation of different regions. Finally, we demonstrate that by transferring a pre-trained deep convolutional neural network (CNN), e.g. VGG19, and fine-tuning the deep CNN using 70% of the real dataset and the semisynthetic data for training, the overall accuracy on the remaining 30% of the real dataset can be eventually improved to 97.76%, which is the highest among all the methods. Our work indicates that the combination of semisynthetic data generation and deep transfer learning is an effective way to improve the accuracy of underwater object classification.

Pole-Like Street Furniture Segmentation and Classification in Mobile LiDAR Data by Integrating Multiple Shape-Descriptor Constraints

Nowadays, mobile laser scanning is widely used for understanding urban scenes, especially for extraction and recognition of pole-like street furniture, such as lampposts, traffic lights and traffic signs. However, the start-of-art methods may generate low segmentation accuracy in the overlapping scenes, and the object classification accuracy can be highly influenced by the large discrepancy in instance number of different objects in the same scene. To address these issues, we present a complete paradigm for pole-like street furniture segmentation and classification using mobile LiDAR (light detection and ranging) point cloud. First, we propose a 3D density-based segmentation algorithm which considers two different conditions including isolated furniture and connected furniture in overlapping scenes. After that, a vertical region grow algorithm is employed for component splitting and a new shape distribution estimation method is proposed to obtain more accurate global shape descriptors. For object classification, an integrated shape constraint based on the splitting result of pole-like street furniture (SplitISC) is introduced and integrated into a retrieval procedure. Two test datasets are used to verify the performance and effectiveness of the proposed method. The experimental results demonstrate that the proposed method can achieve better classification results from both sites than the existing shape distribution method.

MFCSNet: Multi-Scale Deep Features Fusion and Cost-Sensitive Loss Function Based Segmentation Network for Remote Sensing Images

Semantic segmentation of remote sensing images is an important technique for spatial analysis and geocomputation. It has important applications in the fields of military reconnaissance, urban planning, resource utilization and environmental monitoring. In order to accurately perform semantic segmentation of remote sensing images, we proposed a novel multi-scale deep features fusion and cost-sensitive loss function based segmentation network, named MFCSNet. To acquire the information of different levels in remote sensing images, we design a multi-scale feature encoding and decoding structure, which can fuse the low-level and high-level semantic information. Then a max-pooling indices up-sampling structure is designed to improve the recognition rate of the object edge and location information in the remote sensing image. In addition, the cost-sensitive loss function is designed to improve the classification accuracy of objects with fewer samples. The penalty coefficient of misclassification is designed to improve the robustness of the network model, and the batch normalization layer is also added to make the network converge faster. The experimental results show that the classification performance of MFCSNet outperforms U-Net and SegNet in classification accuracy, object details and prediction consistency.

Hyper-Spectral Characteristics in Support of Object Classification and Verification

It is known that a typical hyper-spectral image consists of a sequence of wavelength bands and each wavelength band illustrates a two-dimensional image. This paper presents a HSI classification method including two approaches. The first approach is a probabilistic algorithm for object classification based on the image illustrated by single hyper-spectral wavelength bands, respectively, which generate a sequence of registered object images. Since hyper-spectral images (HSIs) obtained by air-borne or satellite-borne remote sensors has limited size, which may affect the classification accuracy, especially for small objects due to lack of pixels. The second approach investigates the hyper-spectral characteristics for a specific object in terms of uniformed hyper-spectrum energy function in wavelength domain. The hyper-spectral characteristics are applied to object verification, especially for objects which are difficult to distinguish from each other using classic classification methods, so that the accuracy is improved. The classification accuracy for the proposed object-oriented characterization method is evaluated in terms of producer accuracy (PA) and Kappa coefficient based on hyper-spectral images obtained from air-borne or satellite-borne remote sensors, respectively. The numerical results demonstrate that the proposed method can effectively identify and verify objects in hyper-spectral image, especially for those objects that are difficult to be distinguished from each other by classic methods. Furthermore, the proposed object classification method based on hyper-spectral characteristics is compared to conventional methods, including spectral information divergence (SID) method and multiple spectral angle mapper-Markov random fields (MSAM-MRF) method. It can be seen that the accuracy of object classification achieved by the proposed method can be up to 20% higher than that of classic methods.

Object Classification Using CNN-Based Fusion of Vision and LIDAR in Autonomous Vehicle Environment

This paper presents an object classification method for vision and light detection and ranging (LIDAR) fusion of autonomous vehicles in the environment. This method is based on convolutional neural network (CNN) and image upsampling theory. By creating a point cloud of LIDAR data upsampling and converting into pixel-level depth information, depth information is connected with Red Green Blue data and fed into a deep CNN. The proposed method can obtain informative feature representation for object classification in autonomous vehicle environment using the integrated vision and LIDAR data. This method is also adopted to guarantee both object classification accuracy and minimal loss. Experimental results are presented and show the effectiveness and efficiency of object classification strategies.

A New 3D Object Pose Detection Method Using LIDAR Shape Set.

In object detection systems for autonomous driving, LIDAR sensors provide very useful information. However, problems occur because the object representation is greatly distorted by changes in distance. To solve this problem, we propose a LIDAR shape set that reconstructs the shape surrounding the object more clearly by using the LIDAR point information projected on the object. The LIDAR shape set restores object shape edges from a bird’s eye view by filtering LIDAR points projected on a 2D pixel-based front view. In this study, we use this shape set for two purposes. The first is to supplement the shape set with a LIDAR Feature map, and the second is to divide the entire shape set according to the gradient of the depth and density to create a 2D and 3D bounding box proposal for each object. We present a multimodal fusion framework that classifies objects and restores the 3D pose of each object using enhanced feature maps and shape-based proposals. The network structure consists of a VGG -based object classifier that receives multiple inputs and a LIDAR-based Region Proposal Networks (RPN) that identifies object poses. It works in a very intuitive and efficient manner and can be extended to other classes other than vehicles. Our research has outperformed object classification accuracy (Average Precision, AP) and 3D pose restoration accuracy (3D bounding box recall rate) based on the latest studies conducted with KITTI data sets.

Design of Computer Vision System for Objects Recognition in Automation Industries

The field of machine vision has been developing at quick pace. The development in this field, dissimilar to most settled fields, has been both in expansiveness and profundity of ideas and procedures. Object recognition is widely used in the manufacturing industry for the purpose of inspection. Mechanically manufactured parts have recognition difficulties due to manufacturing process including machine malfunctioning, tool wear, and variations in raw material. This paper considers the problem of recognizing and classifying the objects of such parts. RGB images of different objects are used as an input. The Fourier descriptor technique is used for recognition of objects. Artificial Neural Network (ANN) is used for classification of different objects. These objects are kept in different orientations for invariant rotation, translation and scaling. Invariant example acknowledgment utilizing neural systems is an especially appealing methodology on account of its closeness with natural frameworks. This paper shows the effect of different network architecture and numbers of hidden nodes on the classification accuracy of objects.

Development of machine vision system for automated mechanical objects recognition

Object recognition is a type of pattern recognition. Object recognition is widely used in the manufacturing industry for the purpose of inspection. Machine vision techniques are being applied in areas ranging from medical imaging to remote sensing, industrial inspection to document processing, and nanotechnology to multimedia databases. In this paper, mechanical objects recognition used in manufacturing industry is taken into consideration. Mechanically manufactured parts have recognition difficulties due to manufacturing process including machine malfunctioning, tool wear, and variations in raw material. This paper considers the problem of recognising and classifying of such objects. These objects are kept in different orientations for invariant rotation, translation and scaling. The features are extracted using Fourier descriptor technique. Artificial neural network (ANN) with back-propagation learning algorithm is used to train the network and for classification of five different objects. This paper shows the effect of learning rate and momentum on the classification accuracy of objects.

Discriminative Deep Feature Learning for Semantic-Based Image Retrieval

Semantic-based image retrieval plays an important role in many practical applications, which aims to look for images with similar contents. Extracting discriminative representations of images is the real crux of this task. Directly utilizing the results of fully connected layers of convolutional neural networks (CNNs) is one of the best feature extraction methods. However, the fully connected layer is only supervised by the softmax loss, which only aims to maximize the accuracy of object classification and hardly pays attention to the spatial distribution of features, especially the intra-class and inter-class feature distances which are of great importance in semantic-based image retrieval. To compensate the performance degradation due to this reason, we try to address the spatial distribution of the features by two different loss functions. The first loss function jointly combines the softmax loss and the center loss in the training of the CNNs, which simultaneously ensures that the features of images with different contents are separable and the features of images in the same class are close. The second loss function is defined as an improved center loss which not only penalizes the distance between the obtained deep feature and the feature center of its own class, but also encourages a far distance between the feature and the feature centers of any other classes. We have conducted experiments on both ILSVRC data set and Caltech256 data set, demonstrated that the deep features got by our approaches can achieve better performance than other methods in semantic-based image retrieval, and the improved center loss can further outperform the joint supervision of the softmax loss and the center loss.

Top-k multi-class SVM using multiple features

Recent studies have demonstrated the advantages of fusing multi-modal features in improving the accuracy of visual object classification. However, regarding a complex classification task with a large number of categories, previous studies on multiple feature fusion are prone to failure resulting from the occurrence of class ambiguity. In this paper, we address this issue by allowing k (k ≥ 2) guesses at the top instead of only considering the one with the largest prediction score in the framework of multi-view learning. This strategy relaxes the penalty for making an error in the top-k predictions, which can mitigate the challenge of class ambiguity to some extent. To fuse multiple features effectively, we introduce an adaptive weight for each view and exploit an efficient alternating optimization algorithm to learn the optimal classifiers and their corresponding weights jointly. Extensive experiments on several benchmark datasets illustrate the effectiveness and superiority of the proposed model over the state-of-the-art approaches.

3D shape representation with spatial probabilistic distribution of intrinsic shape keypoints

The accelerated advancement in modeling, digitizing, and visualizing techniques for 3D shapes has led to an increasing amount of 3D models creation and usage, thanks to the 3D sensors which are readily available and easy to utilize. As a result, determining the similarity between 3D shapes has become consequential and is a fundamental task in shape-based recognition, retrieval, clustering, and classification. Several decades of research in Content-Based Information Retrieval (CBIR) has resulted in diverse techniques for 2D and 3D shape or object classification/retrieval and many benchmark data sets. In this article, a novel technique for 3D shape representation and object classification has been proposed based on analyses of spatial, geometric distributions of 3D keypoints. These distributions capture the intrinsic geometric structure of 3D objects. The result of the approach is a probability distribution function (PDF) produced from spatial disposition of 3D keypoints, keypoints which are stable on object surface and invariant to pose changes. Each class/instance of an object can be uniquely represented by a PDF. This shape representation is robust yet with a simple idea, easy to implement but fast enough to compute. Both Euclidean and topological space on object’s surface are considered to build the PDFs. Topology-based geodesic distances between keypoints exploit the non-planar surface properties of the object. The performance of the novel shape signature is tested with object classification accuracy. The classification efficacy of the new shape analysis method is evaluated on a new dataset acquired with a Time-of-Flight camera, and also, a comparative evaluation on a standard benchmark dataset with state-of-the-art methods is performed. Experimental results demonstrate superior classification performance of the new approach on RGB-D dataset and depth data.

A support vector machine classifier based on a new kernel function model for hyperspectral data

The kernel function is a key factor to determine the performance of a support vector machine (SVM) classifier. Choosing and constructing appropriate kernel function models has been a hot topic in SVM studies. But so far, its implementation can only rely on the experience and the specific sample characteristics without a unified pattern. Thus, this article explored the related theories and research findings of kernel functions, analyzed the classification characteristics of EO-1 Hyperion hyperspectral imagery, and combined a polynomial kernel function with a radial basis kernel function to form a new kernel function model (PRBF). Then, a hyperspectral remote sensing imagery classifier was constructed based on the PRBF model, and a genetic algorithm (GA) was used to optimize the SVM parameters. On the basis of theoretical analysis, this article completed object classification experiments on the Hyperion hyperspectral imagery of experimental areas and verified the high classification accuracy of the model. The experimental results show that the effect of hyperspectral image classification based on this PRBF model is apparently better than the model established by a single global or local kernel function and thus can greatly improve the accuracy of object identification and classification. The highest overall classification accuracy and kappa coefficient reached 93.246% and 0.907, respectively, in all experiments.

Improving Rough Set Rule-Based Classification by Supplementary Rules

In rough set approaches, decision rules are induced from a given data set consisting of attribute values and a decision value. Induced rules are used to classify new objects, but this classification is not perfect, perhaps because the given data set does not include all possible patterns. No induced decision rules are matched totally for objects having missing patterns, and partially matched decision rules are used to estimate their classes. The classification accuracy of such an object is usually lower than that of an object totally matching decision rules. To improve the classification accuracy, we propose adding supplementary rules to the induced rules, defining the supplementary rules to improve the classification accuracy of objects only partially matching decision rules. We propose an algorithm for inducing supplementary rules, considering four classifiers consisting of supplementary rules together with originally induced rules.We investigate their performance. We also compare their classification accuracies to that of conventional classifier with originally induced rules.

Improving reliability for classification of metallic objects using a WTMD portal

In this paper, a walk-through metal detection (WTMD) portal is used for classification of metallic objects. The classification is based on the inversion of the magnetic polarisability tensor (tensor) of the object. The nature of bias and noise components in the tensor are examined by using real walk-through data, and consequently, a novel classifier is introduced. Furthermore, a novel method for detecting poorly inverted tensors is presented, enabling self-diagnostics for the WTMD portal. Based on the results, the novel methods increase the accuracy of metal object classification and have the potential to improve the reliability of a WTMD system.

Gradient and Laplacian-Based Hyperspectral Anisotropic Diffusion

To improve accuracy and efficiency of object detection and classification with hyperspectral imagery (HSI), we propose a novel smoothing algorithm by coupling of a Laplacian-based reaction term to a classical divergence-based anisotropic diffusion partial differential equation (PDE). In addition, an adaptive parameter is introduced to regularize this nonlinear reaction-diffusion PDE by explicitly integrating the interband correlations with the noise level of each band in HSI. It is also well-known that the interband correlations can be implicitly embedded into the diffusion coefficient of the divergence-based PDE, to allow a selective smoothing that reduces the local homogeneous area variability while preventing smoothing across edges. Therefore, the interband correlations in HSI are exploited in the proposed method in both explicit and implicit ways. As a result, our algorithm is more effective at controlling the behavior of the diffusion evolution when compared to previous multi/hyperspectral diffusion algorithms. The simulations based on both synthetic data and real hyperspectral remote sensing data show that our algorithm can improve the hyperspectral data quality in terms of both visual inspection and image quality indices.

マルチモーダルLDAとNPYLMを用いたロボットによる物体概念と言語モデルの相互学習

Humans develop their concept of an object by classifying it into a category, and acquire language by interacting with others at the same time. Thus, the meaning of a word can be learnt by connecting the recognized word and concept. We consider such an ability to be important in allowing robots to flexibly develop their knowledge of language and concepts. Accordingly, we propose a method that enables robots to acquire such knowledge. The object concept is formed by classifying multimodal information acquired from objects, and the language model is acquired from human speech describing object features. We propose a stochastic model of language and concepts, and knowledge is learnt by estimating the model parameters. The important point is that language and concepts are interdependent. There is a high probability that the same words will be uttered to objects in the same category. Similarly, objects to which the same words are uttered are highly likely to have the same features. Using this relation, the accuracy of both speech recognition and object classification can be improved by the proposed method. However, it is difficult to directly estimate the parameters of the proposed model, because there are many parameters that are required. Therefore, we approximate the proposed model, and estimate its parameters using a nested Pitman--Yor language model and multimodal latent Dirichlet allocation to acquire the language and concept, respectively. The experimental results show that the accuracy of speech recognition and object classification is improved by the proposed method.

The impact of object size on the thematic accuracy of landcover maps

We recently completed the accuracy assessment of a Landsat-derived landcover polygon layer covering the entire province of Alberta (660,000 km2), Canada, for which we gathered reference information for nearly 5000 randomly selected polygons ranging from two hectares to thousands of hectares in size. This gave us the unique opportunity to quantify, for the first time, how the probability of correctly classifying a landcover object varies with its size. Irrespective of whether they are represented as polygons or as sets of connected pixels with the same label, the classification accuracy of landcover objects decreases as their size decreases, steadily for large and medium sizes, and more dramatically when they are within two orders of magnitude of the pixel size of the input image. We show that this size-dependency is bound to occur whenever the size distribution of landcover objects follows an inverse power law. Our results are consistent with previous studies on related issues, confirm the need to account for size when assessing the accuracy of object-based landcover maps, and cast doubts on the validity of (1) recently proposed object-based accuracy estimators, and (2) landscape pattern analyses where the minimum patch size is close to the pixel size.

Object Classification Using Substance Based Neural Network

Object recognition has shown tremendous increase in the field of image analysis. The required set of image objects is identified and retrieved on the basis of object recognition. In this paper, we propose a novel classification technique called substance based image classification (SIC) using a wavelet neural network. The foremost task of SIC is to remove the surrounding regions from an image to reduce the misclassified portion and to effectively reflect the shape of an object. At first, the image to be extracted is performed with SIC system through the segmentation of the image. Next, in order to attain more accurate information, with the extracted set of regions, the wavelet transform is applied for extracting the configured set of features. Finally, using the neural network classifier model, misclassification over the given natural images and further background images are removed from the given natural image using the LSEG segmentation. Moreover, to increase the accuracy of object classification, SIC system involves the removal of the regions in the surrounding image. Performance evaluation reveals that the proposed SIC system reduces the occurrence of misclassification and reflects the exact shape of an object to approximately 10–15%.

Decoding of Repeated Objects from Local Field Potentials in Macaque Inferior Temporal Cortex

Stimulus repetition produces a decrease of the response and affects neuronal synchronization of macaque inferior temporal (IT) neurons. Previously we showed that such stimulus-specific adaptation results in a decreased accuracy by which IT neurons encode repeated compared to non-repeated objects. Not only spiking activity, but also local field potentials (LFPs) are affected by repetition. Here we ask how the repetition-induced changes in IT LFPs affect object decoding accuracy. To answer this, we recorded local field potentials using a laminar microelectrode in macaque IT. We presented two familiar stimuli each for 500 ms successively with an inter-stimulus interval of 500 ms. Trials consisted either of a repetition of the same stimulus or of their alternation. Machine learning-based classifier was employed to decode stimulus identity from the LFP power in different frequency bands of each penetration. We found that the object classification accuracy depended strongly on spectral frequency, with frequencies below 30 Hz (alpha and beta) producing greater accuracies than gamma bands. However, the effect of repetition on classification accuracy was stronger at the gamma frequencies, showing a decrease in classification accuracy for repeated stimuli and a tendency for an improved object encoding when the stimulus was preceded by a different stimulus. The present results demonstrate that due to adapting input, stimulus encoding in IT (1) can be more accurate for stimuli that differ from recently preceding ones while being impaired for stimuli that are repeated, and (2) these effects are more pronounced at high spectral frequencies of the LFP.