Artificial intelligence is widely applied to automate Diabetic retinopathy diagnosis. Diabetes-related retinal vascular disease is one of the world’s most common leading causes of blindness and vision impairment. Therefore, automated DR detection systems would greatly benefit the early screening and treatment of DR and prevent vision loss caused by it. Researchers have proposed several systems to detect abnormalities in retinal images in the past few years. However, Diabetic Retinopathy automatic detection methods have traditionally been based on hand-crafted feature extraction from the retinal images and using a classifier to obtain the final classification. DNN (Deep neural networks) have made several changes in the previous few years to assist overcome the problem mentioned above. We suggested a two-stage novel approach for automated DR classification in this research. Due to the low fraction of positive instances in the asymmetric Optic Disk (OD) and blood vessels (BV) detection system, preprocessing and data augmentation techniques are used to enhance the image quality and quantity. The first step uses two independent U-Net models for OD (optic disc) and BV (blood vessel) segmentation. In the second stage, the symmetric hybrid CNN-SVD model was created after preprocessing to extract and choose the most discriminant features following OD and BV extraction using Inception-V3 based on transfer learning, and detects DR by recognizing retinal biomarkers such as MA (microaneurysms), HM (hemorrhages), and exudates (EX). On EyePACS-1, Messidor-2, and DIARETDB0, the proposed methodology demonstrated state-of-the-art performance, with an average accuracy of 97.92%, 94.59%, and 93.52%, respectively. Extensive testing and comparisons with baseline approaches indicate the efficacy of the suggested methodology.

In practical industrial application, the fault samples collected from rotating machinery are frequently unbalanced, which will create difficulties when it comes to diagnosis. Besides, the variation of working conditions and noise factors will further reduce the diagnosis’s accuracy and stability. Considering the above problems, we established a model based on deep Wasserstein generative adversarial network with gradient penalty (DWGANGP). In this model, the unbalanced fault data set will first be trained by the sample generation network to generate synthetic samples, which will be used to restore the balance. A one-dimensional convolutional neural network with a specific structure is then used as the fault diagnosis network to classify the reconstructed equilibrium samples. The experimental results show that the proposed sample generation network can generate high-quality synthetic samples under highly imbalanced data, and the diagnostic network has a fast training convergence. Compared to the combination methods of support vector machines, back propagation neural network and deep belief network, our method has a 74% average accuracy in all unbalanced experimental conditions, which has 64%, 69% and 87% averages leading, respectively.

The paradigm of “deep-learning visual perception + hand–eye transformation + motion planning” for robot grasping has demonstrated viable capabilities in specific scenarios. However, its further development faces challenges in handling complex and dynamic environments. This article proposes a keypoint detection network-driven visual servoing grasping framework. First, we develop an efficient two-stage keypoint detector to perform real-time inference of sparse image-plane features for the target. Subsequently, a low-pass filtering algorithm is employed to smoothen the detected keypoints. These processed keypoints are then used in an image-based visual servoing (IBVS) controller to calculate the robot joint velocities, enabling precise tracking. A specialized dataset for training and evaluation was constructed using domain randomization techniques, comprising 11 K samples across six categories. Comprehensive experiments demonstrate the detector’s low latency and accurate performance, even in low lighting, overexposure, partial occlusion, and densely packed environments. Static and dynamic grasping experiments validate that this framework achieves localization accuracy superior to five pixels and an overall grasping success rate exceeding 70% under unknown hand–eye calibration. The dataset is provided at github.com/hijunqi/VS_grasping_keypoint_detection_dataset.

Predicting the time series of 10.7-cm solar radio flux is a challenging task because of its daily variability. This paper proposed a non-linear method, a convolutional and recurrent neural network combined model to achieve end-to-end F10.7 forecasts. The network consists of a one-dimensional convolutional neural network and a long short-term memory network. The CNN network extracted features from F10.7 original data, then trained the feature signals in the long short-term memory network, and outputted the predicted values. The F10.7 daily data during 2003–2014 are used for the testing set. The mean absolute percentage error values of approximately 2.04%, 2.78%, and 4.66% for 1-day, 3-day, and 7-day forecasts, respectively. The statistical results of evaluating the root mean square error, spearman correlation coefficient shows a superior effect as a whole for the 1–27 days forecast, compared with the ordinary single neural network and combination models.

Effective multi-agent teamwork can be facilitated by using personas to decompose goals into lower-level team subtasks through a shared understanding of multi-agent tasks. However, traditional methods for role discovery and assignment are not scalable and fail to adapt to dynamic changes in the environment. To solve this problem, we propose a new framework for learning dynamic role discovery and assignment. We first introduce an action encoder to construct a vector representation for each action based on its characteristics, and define and classify roles from a more comprehensive perspective based on both action differences and action contributions. To rationally assign roles to agents, we propose a representation-based role selection policy based on consideration of role differences and reward horizons, which enables agents to play roles dynamically by dynamically assigning agents with similar abilities to play the same role. Agents playing the same role share their learning of the role, and different roles correspond to different action spaces. We also introduce regularizers to increase the differences between roles and stabilize training by preventing agents from changing roles frequently. Role selection and role policy integrate action representations and role differences in a restricted action space, improving learning efficiency. We conducted experiments in the SMAC benchmark and showed that our method enables effective role discovery and assignment, outperforming the baseline on four of the six scenarios, with an average improvement in win rate of 20 $$\%$$% , and is effective in hard and super hard maps. We also conduct ablation experiments to demonstrate the importance of each component in our approach.