In order to solve the problem of difficulty in learning cooperative strategies and decision-making of multi-agent caused by the incomplete information and the inability to guarantee stable communication in the partial observable multi-agent adversarial environment, inspired by the learning and reasoning functions of the human cerebral cortex through memory, a new multi-agent distributed cooperative strategy learning method based on incomplete information prediction in a partially observable adversarial environment is proposed. Through the use of historical memory and current observation information, support vector regression (SVR) is used to predict invisible information in the environment, and the prediction information of the unobserved part and the observed information are fused as the basis for cooperative strategy learning and decision-making. Then, through the use of decentralized multi-agent reinforcement learning method for cooperative strategy learning, the decision model of each agent in the team is obtained. The method is verified by multi-agent cooperative strategy learning and decision-making in a typical partial observable adversarial environment. The results show that this method can significantly improve the level of multi-agent cooperative decision-making while ensuring high prediction accuracy for unobserved parts.
Non-contact damage detection of grotto murals is a challenging task, due to their minor cracks and subtle defects. In this paper, a lightweight neural network-based grotto mural damage detection algorithm, named Ghost-C3SE YOLOv5, is proposed, which can effectively reduce the network size while obtaining reasonable damage detection results. First, based on the YOLOv5 detection algorithm, the dimension of the convolution layer is reduced, by adjusting the network structure and integrating a Ghost module. Second, the channel attention mechanism is introduced into the feature extraction backbone network, to adjust the weight of each feature according to its importance, which can accelerate the convergence speed of the loss function during the model training process. Finally, the experiment results have shown that the lightweight model Ghost-C3SE YOLOv5, applied on the Pascal Voc dataset, reduces the number of parameters by about 22.55 million while ensuring the detection precision and recall rate. Also, the training time and the model size are reduced by 36.21% and 46.04%, respectively. Precision is increased by 1.29% and recall remains comparative, while the utilization rate of the GPU is improved by 41.55%. This has addressed the shortcomings of laborious process and low accuracy in manual detection of the grotto murals. Furthermore, a real-time detection performance of 44.86 FPS is achieved here for damage detection of murals in the YunGang Grottoes, which has overcome the drawbacks of high model complexity, high computational cost, slow detection speed and low memory utilization of various existing deep learning algorithms.
In order to solve the problems of limited computing resources on the fuselage and autonomous recognition of visual images during the real-time control of the flapping-wing flying robot, and to realize the intelligent navigation control of the flapping-wing flying robot based on vision, this paper proposes a visual image processing method and edge detection method for a flapping-wing flying robot based on cluster analysis. Using cluster analysis method to classify the visual image pixel data of the flapping-wing flying robot, and using the classic edge detection algorithm to perform edge detection on the flying robot image after clustering analysis, and reasonably extracting the visual target contour features and environmental edges of the flapping-wing flying robot feature. An experimental platform for a flapping-wing flying robot was built with a flapping-wing flying robot and an airborne vision camera. Flight experiments are carried out on the visual image processing method and image edge detection method of flying robot based on cluster analysis. The experimental results verify the feasibility and effectiveness of the method in this paper. The innovation of this paper is embodied in the application of cluster analysis method to the visual image processing of the flapping-wing flying robot, and the edge detection of the visual image of the flapping-wing flying robot.
针对并行深度卷积神经网络算法在大数据环境下存在冗余特征计算过多、卷积运算性能不足和参数并行化合并效率低等问题,提出了基于Winograd卷积的并行深度卷积神经网络优化算法.首先,该算法提出基于余弦相似度与归一化互信息的特征过滤策略,通过先筛选后融合的方式消除了通道间对于冗余特征的计算,以此解决了冗余特征计算过多的问题;然后,结合MapReduce提出了并行Winograd卷积策略,通过使用并行化Winograd卷积运算替换传统卷积运算的方式来提升卷积运算的性能,以此解决了卷积运算性能不足的问题;最后,提出基于任务迁移的负载均衡策略,通过动态负载迁移的方式来均衡集群中各节点之间的负载,降低了集群总体的平均反应时长,以此解决了参数并行化合并效率低的问题.实验表明,WP-DCNN算法显著降低了 DCNN在大数据环境下的训练代价,而且对并行DCNN的训练效率也有大幅提升.
针对基于卷积神经网络的去雨方法感受野受限的问题,结合Swin Transformer和卷积神经网络各自的优势,提出了 一种融合全局和局部特征的单幅图像去雨方法.首先通过卷积神经网络对图像的局部特征进行初步提取;其次通过基于Swin Transformer的多支路网络对不同特征空间内的全局信息进行学习;最后将提取出的多支路全局特征与局部特征进行融合,实现无雨图像的恢复.在多个数据集上与多种主流单幅图像去雨方法进行了对比实验.结果表明,所提方法生成的结果在峰值信噪比和结构相似性指标上都具有一定优势,验证了所提方法在图像去雨任务上的有效性.
