International Journal of Pattern Recognition and Artificial Intelligence

In this paper, a two-stage multi-category recognition structure based on texture features is proposed. This method can solve the problem of the decline in recognition accuracy in the scene of lightweight training samples. Besides, the problem of recognition effect different in the same recognition structure caused by the unsteady iris can also be solved. In this paper’s structure, digitized values of the edge shape in the iris texture of the image are set as the texture trend feature, while the differences between the gray values of the image obtained by convolution are set as the grayscale difference feature. Furthermore, the texture trend feature is used in the first-stage recognition. The template category that does not match the tested iris is the elimination category, and the remaining categories are uncertain categories. Whereas, in the second-stage recognition, uncertain categories are adopted to determine the iris recognition conclusion through the grayscale difference feature. Then, the experiment results using the JLU iris library show that the method in this paper can be highly efficient in multi-category heterogeneous iris recognition under lightweight training samples and unsteady state.

Power consumption attacks monitoring on artificial intelligence (AI) chips play a critical role in the vehicular AI systems. However, most of the current monitoring and management methods focus on the trustworthiness of industrial equipment instead of resource-constrained edge devices. To address the above problem, a closed-loop module for monitoring and management of vehicular AI chips based on fitting and filtering to resist power consumption attacks is proposed in this paper. First, considering the characteristics of power, we propose a raw data correction approach for power monitoring to monitor abnormal power consumption. Second, we address the challenging problem of precision temperature monitoring to monitor the abnormal temperature of the chip, especially in a wide temperature range. Finally, the established method is applied to attack surveillance and transformed into a power consumption management problem solved by dynamic voltage and frequency scaling (DVFS) technology. As the experimental results reveal, compared with existing methods of power and temperature monitoring and power consumption control in wide temperature, our method can achieve significantly improved monitoring and managing performance.

Aspect-level sentiment analysis aims to predict the sentiment polarity toward a specific aspect in a sentence. Most current approaches are based on deep learning and the attention mechanism. However, these models cannot simultaneously include the context semantic information carried by local words and the global syntactic information possibly carried by remote words. In this paper, we propose a local semantic and global syntactic integration scheme, which employs a local focus mechanism over local context words and exploits improved graph convolutional networks over dependency tree to encode global syntactic information. Moreover, multi-head attention is used to capture both the semantic information and also the interactive information between semantics and syntactic features. Experimental results on five datasets show the effectiveness of our model over a series of latest models.

Aiming to obtain more discriminative features in scene images and overcome the impacts of intra-class differences and inter-class similarities, the paper proposes a scene recognition method that combines attention and context information. First, we introduce the attention mechanism and build a multi-scale attention model. Discriminative information considers salient objects and regions by means of channel attention and spatial attention. Besides, the central loss function joint supervision strategy is introduced to further reduce the misjudgment of intra-class differences. Second, a model based on multi-level context information is proposed to describe the positional relationship between objects, which can effectively alleviate the influence of the similarity of objects between classes. Finally, the two models are merged to give full play to the compatibility of features, so that the final feature representation not only focuses on the effective discriminant information, but also manifests the relative position relationship between significant objects. Extensive experiments have proved that the method in this paper effectively solves the problem of insufficient feature representation in scene recognition tasks, and improves the accuracy of scene recognition.

Temporal action detection is one of the most important and challenging tasks in video analysis. Due to its wide application prospects, it has received extensive attention in recent years. With the development of deep learning, great progress has been made in temporal behavior detection, but there are still many difficulties to be solved, such as accurate proposal generation and high computational cost. In this paper, deep learning-based temporal action detection methods are classified according to full supervision and weak supervision, and then the representative models of the two methods are summarized in detail, and the ideas, advantages and disadvantages of different models and the evolution between different models are analyzed. At the same time, the performance of different models on mainstream datasets is compared. The mainstream dataset and evaluation index used in temporal action detection are introduced in detail, and the calculation method of evaluation index is also elaborated. Finally, through in-depth analysis, the possible future research directions of temporal action detection and the whole review are summarized.

Compared with printed Chinese characters, handwritten Chinese characters are affected by writing coherence and anthropic factors, resulting in differences in stroke deformation, displacement, tilt, length, and thickness. This consequently affects the recognition of handwritten Chinese characters. A method based on generative adversarial networks (GANs) of Chinese character handwriting-to-printing font conversion is proposed herein, where Chinese handwriting-to-printing font conversion is represented as a font style conversion and regarded as the optimal state of normalization of handwritten Chinese characters. First, an encoder–decoder is integrated into a generative adversarial network, and a symmetric network extracts handwritten font multi-scale information. Subsequently, U-Net with skip connection is used to extract deep features and reduce the large amount of low-level information shared between the input and output. Furthermore, the skip connection reduces information loss during down-sampling. Finally, the integration loss measures the differences in the character structure and font style between the generated Chinese character image and the target font image by combining font style, encoding consistency, and $L_1$ losses. An experiment to convert Chinese character handwriting to multiple printing fonts is performed on the CASIA and CASIA-HWDB1.1 datasets, in which image pixel difference and character recognition accuracy are used as evaluation metrics. The obtained results show a more normalized visual writing style and contribute to recognition improvement, thereby verifying the effectiveness of this method.

Machine learning techniques in medical imaging systems are accurate, but minor perturbations in the data known as adversarial attacks can fool them. These attacks make the systems vulnerable to fraud and deception, and thus a significant challenge has been posed in practice. We present the gradient-free trained sign activation networks to detect and deter adversarial attacks on medical imaging AI systems. Experimental results show that a higher distortion value is required to attack our proposed model than the other existing state-of-the-art models on MRI, Chest X-ray, and Histopathology image datasets, where our model outperforms the best and is even twice superior. The average accuracy of our model in classifying the adversarial examples is 88.89%, whereas those for MLP and LeNet are 81.48%, and that of ResNet18 is 38.89%. It is concluded that the sign network is a solution to defend adversarial attacks due to high distortion and high accuracy on transferability. Our work is a significant step towards safe and secure medical AI systems.

Image inpainting removes unwanted objects from the image, signifying the original image restoration. Even though several techniques are introduced for image inpainting, but still, there are several challenging issues associated with the conventional methods regarding data loss, which are effectively handled based on the proposed approach. In this paper, we propose an effective hybrid image inpainting method that is termed as ALGDKH, which is the hybridization of Ant Lion–Gray Wolf Optimizer (ALG)-based Markov random field (MRF) modeling, deep learning, $K$-nearest neighbors (KNN) and the harmonic functions. The crack input image is forwarded as an input to Markov random field modeling to obtain image inpainting, where the MRF energy is minimized based on the ALG. Then, the same crack image is subjected to the Whale–MBO-based DCNN, KNN with Bhattacharya distance and Bi-harmonic function modules to obtain the inpainting results. Finally, the results from the proposed ALG-based Markov random field modeling, Whale–MBO-based DCNN, KNN with Bhattacharya distance and Bi-harmonic function modules are fused through Bayes-probabilistic fusion for the final inpainting results. The proposed method produces the maximal PSNR of 38.14dB, maximal SDME of 75.70dB and the maximal SSIM of 0.983.

Siamese network-based object tracking algorithms have recently gained popularity due to continuous improvement in tracking accuracy and robustness. However, these trackers are often limited by inefficient initialization and unstable online update. The model drift accumulates as more frames are processed. To address the above issues, this paper introduced a method that combines both deep learning and conventional regression algorithms (SVR, Support Vector Regression). Specifically, we propose a feature projection algorithm, which can effectively reduce the dimension of the features extracted from deep neural networks and improve the distinguishability of them at the same time. To make the features more robust for SVR training, we further propose two feature aggregation methods at both the channel level and the spatial level. We update the SVR model online instead of the deep neural network to make the tracking process more robust. Extensive experiments on four challenging benchmarks indicate that our proposed tracker is superior to baseline methods both qualitatively and quantitatively.

In this paper, we propose a computationally efficient method for traffic sign detection. Our methodology uses deep learning networks introduced for semantic image segmentation and classification of region of interests generated by image segmentation. The region of interests are detected by segmenting the road scene image dataset and hypothesis tests are applied to accept or reject the probability of being a traffic sign in the detected region of interest. For hypothesis tests, color feature, and the location of the traffic sign with respect to the segmented road information is used. The main contribution of our method is the introduction of hypothesis tests that mutually couple two deep learning networks trained by well-known datasets publicly available for benchmarking purposes. To test and evaluate our traffic sign detection system, we use the German traffic sign detection benchmark dataset with a large set of traffic signs and during its training, The cityscapes dataset offering labeled urban scenes is also leveraged by the traffic sign detection system. Our experimental results illustrate the effectiveness performance metrics are reached at 90.81%, 94.76% and 92.74% in precision, recall and F-measure, respectively. The runtime cost is around 0.4s for an image on an ordinary laptop computer.

With the development of modern agricultural facilities, crop diseases recognition, nutritional status and morphology achieved rapid growth. To avoid yield loss caused by the delay of disease detection, digital images that contain information with respect to crop growth, disease type and nutrition deficiency have been studied by some researchers. However, traditional image processing methods fail to extract typical disease features of crop images with ambiguous disease information. In this paper, a crop disease image recognition technique based on the salient region and hierarchical indexing was proposed. Improved Harris algorithm and maximum radius were used to calculate the widest salient region. In order to eliminate the effect of different salience distribution ranges between different features, a group of images in the cucumber disease image library were normalized. Experiment results indicate that the time complexity of each algorithm will go up as the size of the dataset increase. Especially when testing large datasets, nonhierarchical and nonclustering, hierarchical and nonclustering and hierarchical based on points all tend to raise the algorithm’s time complexity. Plant Village dataset and AI Challenger 2018 dataset were utilized to compare the recognition performances among the models based on machine learning, neural network, deep learning and our methods. The experiment results show that the method proposed in this paper is capable of recognizing local similar images effectively rather than global similar images, therefore, it has better recognition performance than the model learning methods in the early detection stage of crop disease.

Nonnegative matrix factorization (NMF), distinguished from the approaches for holistic feature representation, is able to acquire meaningful basis images for parts-based representation. However, NMF does not utilize the data-label information and usually achieves undesirable performance in classification. To address the above-mentioned problem of NMF, this paper proposes a new enhanced NMF (ENMF) method for facial image representation and recognition. We seek to learn powerfully discriminative feature by a label-based regularizer which describes the relationship between the data. It is desired that minimizing the regularizer makes the data from the same class have high similarity and the data from different classes have low similarity. This good property will contribute to improving the performance of NMF. Therefore, we propose an objective function of ENMF by incorporating the label-based regularizer into the loss function. Subsequently, we find the stationary point of the constructed auxiliary function by means of Cardano’s formula and derive the update rules of our ENMF algorithm. The convergence of the proposed ENMF is both theoretically sound and empirically validated. Finally, the proposed ENMF method is successfully applied to face recognition. Four publicly available face datasets, namely AR, Caltech 101, Yale, and CMU PIE, are chosen for evaluations. Compared with the state-of-the-art NMF-based algorithms, the experimental results illustrate that the proposed ENMF algorithm achieves superior performance.

Local search algorithms are widely applied in solving large-scale distributed constraint optimization problem (DCOP). Distributed stochastic algorithm (DSA) is a typical local search algorithm to solve DCOP. However, DSA has some drawbacks including easily falling into local optima and the unfairness of assignment choice. This paper presents a novel local search algorithm named VLSs to solve the issues. In VLSs, sampling according to the probability corresponding to assignment is introduced to enable each agent to choose other promising values. Besides, each agent alternately performs a greedy choice among multiple parallel solutions to reduce the chance of falling into local optima and a variance adjustment mechanism to guide the search into a relatively good initial solution in a periodic manner. We give the proof of variance adjustment mechanism rationality and theoretical explanation of impact of greed among multiple parallel solutions. The experimental results show the superiority of VLSs over state-of-the-art DCOP algorithms.

The intelligent regulation and control strategies for rice and wheat combine harvesters’ operation are lacking and the rule of parameter matching is fuzzy in China, around these issues. The dynamic correlation control law among the parameters of rice and wheat, cleaning operation parameters of combine harvesters, the cleaning loss rate and impurity rate, and so on are studied. The intelligent control model for the rice and wheat combine harvester is established based on case-based reasoning (CBR). According to different rice and wheat varieties, water content and other rice and wheat properties, the control scheme of cleaning fan speed, air distributor plate angle and upper sieve opening with low cleaning impurity rate and cleaning loss rate is provided. Through the development of web-based cleaning intelligent control expert system and experimental evaluation, the feasibility and effectiveness of the CBR method in the intelligent control filed of rice and wheat combine harvesters are verified.

Aiming at the low efficiency and high energy consumption of unmanned ships traversing the entire area, a complete coverage path planning algorithm based on the improved A-star algorithm is proposed. The positioning and vision systems of unmanned ships are used to digitize the actual water information, and the grid method is used to convert the information into an environmental map that can be planned. Compared to the trapezoidal partition of unity method and the short-side reciprocating traversal algorithm in the traversal process, experiments show that path planning is more efficient with the boustrophedon partition of unity method and the long-side reciprocating traversal algorithm. Aiming at the “dead zone”, an improved A-star algorithm is proposed on the basis of the traditional A-star algorithm, that it can shorten about 1/4 path using the proposed algorithm. Simulation shows that the improved A-star algorithm can shorten the traversal path to 40 steps but the traditional A-star algorithm needs 54 steps. Navigation test shows that the proposed algorithm can shorten the traversal path and improve traversal efficiency while ensuring the coverage of unmanned ships.