Target-tools recognition method based on an image feature library for space station cabin service robots

A synthetic aperture radar (SAR) automatic target recognition can effectively improve the utilization efficiency of SAR image data. In order to improve the generalization and accuracy of SAR image target recognition, a convolutional neural network model is proposed to be applied to SAR image target recognition. Firstly, the network structure is designed for the characteristics of SAR image. Secondly, the introduction of dense block network structure improves the generalization performance of the model. Finally, Dropout reduced the computational complexity of the network and improved the generalization performance of the model. Experimental data were obtained from the United States Moving and Stationary Target Acquisition and recognition database. Experimental results of 10 types of target recognition showed that the overall recognition rate of the improved convolutional neural network is 99.18%. The convolutional neural network model proposed in this study improves the accuracy and generalization of the network, and is an effective method for target recognition of SAR images.

Micro-Doppler feature can provide rich information for motion target classification and recognition of radar. Feature selection is a key step in target recognition and it is a valuable work to analyze the recognition features and the influential factors on recognition accuracy. Four features extracted from foot component and torso component are selected to recognize one walking person and two walking persons based on micro-Doppler in this paper. The impact of time window length and signal-to-noise ratio on the recognition accuracy is analyzed. The contribution of single feature to recognition rate is researched too. The corresponding results of simulation are shown in the paper.

This paper proposes a low-resolution ground surveillance radar automatic target recognition(ATR) method based on onedimensional convolutional neural network (1D-CNN), which solves the problem of overfitting using complex CNN for data classification. First, the target recognition algorithm combines the time-domain waveform, power spectrum, and power transform spectrum into the three channels of the established 1D-CNN input. After that, the autoencoder is used to reduce the feature dimension and improve the classifier's ability to select parameters autonomously. Finally, the Bayesian hyperparameter optimization method is used to optimize hyperparameters, which not only simplifies the network structure, but also reduces the parameter calculation scale. We tested our method with the collected data to classify people and cars, and the results showed that the recognition accuracy rate has reached 99%. Compared with the traditional artificial feature extraction target recognition method, our model has better recognition performance and adaptability.

Facial expression recognition has been applied in many fields such as human computer interaction, patient monitoring, neurology, social robot… Although facial expression recognition has gained some encourage results, there are still many challenges such as the change of illumination, blur, etc. Especially, recognizing between sadness and anger expression raises more barriers. In this paper, we proposed a framework using only salient facial regions, but it would be able to improve the accuracy of facial expression recognition. In this framework, one of the most state-of-the-art descriptor, called Pyramid of Local Phase Quantization descriptor (PLPQ) was used to robust with respect to image blur. The experiment achieved 97.7% accuracy recognition rate on the extend Cohn-Canade (CK+) database, and outperformed than other state-of-the-art facial expression recognition methods.

Color-based object recognition

Local invariant feature extraction, as one of the main problems in the field of computer vision, has been widely applied to image matching, splicing and target recognition etc. Lowe’s scale invariant feature transform (known as SIFT) algorithm has attracted much attention due to its invariance to scale, rotation and illumination. However, SIFT is not robust to affine deformations, because it is based on the DoG detector which extracts keypoints in a circle region. Besides, the feature descriptor is represented by a 128-dimensional vector, which means that the algorithm complexity is extremely large especially when there is a great quantity of keypoints in the image. In this paper, a new feature descriptor, which is robust to affine deformations, is proposed. Considering that circles turn to be ellipses after affine deformations, some improvements have been made. Firstly, the Gaussian image pyramids are constructed by convoluting the source image and the elliptical Gaussian kernel with two volatile parameters, orientation and eccentricity. In addition, the two parameters are discretely selected in order to imitate the possibilities of the affine deformation, which can make sure that anisotropic regions are transformed into isotropic ones. Next, all extreme points can be extracted as the candidates for the affine-invariant keypoints in the image pyramids. After accurate keypoints localization is performed, the secondary moment of the keypoints’ neighborhood is calculated to identify the elliptical region which is affineinvariant, the same as SIFT, the main orientation of the keypoints can be determined and the feature descriptor is generated based on the histogram constructed in this region. At last, the PCA method for the 128-dimensional descriptor’s reduction is used to improve the computer calculating efficiency. The experiments show that this new algorithm inherits all SIFT’s original advantages, and has a good resistance to affine deformations; what’s more, it is more effective in calculation and storage requirement.

With the rapid development of electronic computers and information technology, face recognition is widely used in fields such as enterprises, entertainment, information security, and daily life. However, the current face recognition technology is still relatively poor in distinguishing facial features, resulting in a low accuracy of face recognition, which cannot meet increasing application requirements. For this reason, it is necessary to develop more accurate face recognition technology. Residual neural network (ResNet) neural network is a deep residual learning convolutional neural network, which can be used to solve the degeneration problem (i.e., after adding more layers to the neural network, the performance drops rapidly) of deep-learning neural networks. We aim to study the specific principles, calculation methods, and characteristics of the ResNet neural network, analyze the complexity of the classroom environment, propose the use of ResNet network for face feature extraction, and use additive angular margin loss for deep face recognition (ArcFace) as the loss function of the ResNet network to improve the ResNet network. Because the ArcFace function has the advantages of high performance, easy programming, low complexity and high training efficiency. We use this scheme to conduct field tests of face recognition on many students in the classroom. The test results show that this scheme enhances the discrimination of facial features and improves the accuracy of face recognition. In the case of different numbers of people, facial defects, and strong light exposure, the system can still detect and recognize faces stably. In the case of facial defects, the recognition accuracy rate is still 63.3%, and the recognition accuracy rate is still more than 60% under the illumination of strong light, and the recognition accuracy rate under correct conditions is more than 70%.

LocalFace: Learning significant local features for deep face recognition

A fault diagnosis method for the rotating machinery based on improved Convolutional Neural Network (CNN) with Gray-Level Transformation (GLT) is proposed to increase the accuracy of the recognition adopting the multiple sensors. The Symmetrized Dot Pattern (SDP) in this method is applied to fuse the data of the multiple sensors, and the multi-color value method is adopted to increase the feature dimension. The grayscale and GLT are used to reduce the dimension of the SDP image. The SDP grayscale image is finally input to the CNN network for training recognition. The research results show that the diagnosis accuracy of the rolling bearing system based on the novel method is up to 98.6 %. Compared with the method without the multi-color value and GLT, the recognition accuracy of the proposed method is improved by 22.3 %, and the training time is reduced by about one third. The research work reveals that the developed method has the potential application value under the multi-sensor working conditions for the fault diagnosis.

The goal of this research was to study and optimize multi-temporal RGB images obtained by a UAV (unmanned aerial vehicle). A digital camera onboard the UAV allows obtaining data with a high temporal and spatial resolution of ground objects. In the case considered by us, the object of study is agricultural fields, for which, based on numerous images covering the agricultural field, image mosaics (orthomosaics) are constructed. The acquisition time for each orthomosaic takes at least several hours, which imposes a change in the illuminance of each image, when considered separately. Orthomosaics obtained in different periods of the year (several months) will also differ from each other in terms of illuminance. For a comparative analysis of different parts of the field (orthomosaic), obtained in the same time interval or comparison of areas for different periods of time, their alignment by illumination is required. Currently, the majority of alignment approaches rely rather on colour (RGB) methods, which cannot guarantee finding efficient solutions, especially when it is necessary to obtain a quantitative result. In the paper, a new method is proposed that takes into account the change in illuminance during the acquisition of each image. The general formulation of the problem of light correction of RGB images in terms of assessing the colour vegetation index Greenness is considered. The results of processing real measurements are presented.

Passive sonar target recognition is a challenging task due to the complex milieu of the ocean. Most of the state of the art target recognition systems depend on hand engineered feature extraction schemes in order to effectively represent the target signatures, based on expert knowledge. Due to the whimsical nature of the sources and medium, such feature engineering methods often fail to yield invariant features from the observations. In this paper, a deep unsupervised feature learning approach capable of capturing invariant features from the sensory signal stream through multi layered hierarchical abstraction has been adopted. These abstractions learned by the higher layers are mostly invariant and can be used as the discriminative features for the purpose of classification.

Image target detection and recognition is an important research area in image understanding and computer vision, and has attracted extensive attention of scholars for its wide application. In recent years, with the enlightenment of human vision system, deep learning technology has been put forward, and has become a hot spot. The deep learning model has a multi-layer network structure, which can overcome the shortcomings of the traditional image target recognition methods. In the detection and recognition of image targets, deep learning methods are more and more widely used. This is because the deep learning multilayer network structure can express complex functions in a more concise way than the shallow network, and can learn the deep feature representation. Therefore, this paper uses the method based on deep learning, researches on image target detection and recognition, mainly introduces the region proposal based on the depth of learning based target recognition method of support vector machine algorithm and target detection depth. In this paper, deep learning is applied to image detection and recognition, further to improve the accuracy of image detection and recognition. It has important guiding significance in deep learning and image research.

The current dynamic gesture contour feature extraction method has the problems that the recognition rate of dynamic gesture contour feature and the recognition accuracy of dynamic gesture type are low, the recognition time is long, and comprehensive is poor. Therefore, we propose a dynamic gesture contour feature extraction method using residual network transfer learning. Sensors are used to integrate dynamic gesture information. The distance between the dynamic gesture and the acquisition device is detected by transfer learning, the dynamic gesture image is segmented, and the characteristic contour image is initialized. The residual network method is used to accurately identify the contour and texture features of dynamic gestures. Fusion processing weights are used to trace the contour features of dynamic gestures frame by frame, and the contour area of dynamic gestures is processed by gray and binarization to realize the extraction of contour features of dynamic gestures. The results show that the dynamic gesture contour feature recognition rate of the proposed method is 91%, the recognition time is 11.6 s, and the dynamic gesture type recognition accuracy rate is 92%. Therefore, this method can effectively improve the recognition rate and type recognition accuracy of dynamic gesture contour features and shorten the time for dynamic gesture contour feature recognition, and the F value is 0.92, with good comprehensive performance.

Side-scan sonar (SSS) target recognition is an important part of building an underwater detection system and ensuring a high-precision perception of underwater information. In this paper, a novel multi-channel multi-location attention mechanism is proposed for a multi-modal phased transfer side-scan sonar target recognition model. Optical images from the ImageNet database, synthetic aperture radar (SAR) images and SSS images are used as the training datasets. The backbone network for feature extraction is transferred and learned by a staged transfer learning method. The head network used to predict the type of target extracts the attention features of SSS through a multi-channel and multi-position attention mechanism, and subsequently performs target recognition. The proposed model is tested on the SSS test dataset and evaluated using several metrics, and compared with different recognition algorithms as well. The results show that the model has better recognition accuracy and robustness for SSS targets.

Automatic target recognition (ATR) is of increasing importance for the modern radar system, where the high-resolution range profile (HRRP) is essential. However, the recognition accuracy and sensitivity to dataset should be optimized for practical applications. Herein, we proposed a novel algorithm for HRRP target recognition based on a convolutional capsule network rather than neurons in conventional deep neural networks. The capsules were vectors trained to learn latent features from input HRRP, with the length of the vector representing the confidential probability. The convolution dynamic routing mechanism was applied between capsule layers by shared transformation matrices and constrained routing procedures in local kernels, reducing the size of parameters and computational expense. Experiments on measured data proved that the proposed algorithm outperforms other existing methods with higher recognition accuracy and less sensitivity to training size. This study provided a promising and effective approach for HRRP target recognition.