We are embracing the fourth wave of human development, which is a critical transition from the information society to an intelligent society integrating human beings, the physical world, and the cyberspace. In recent years, computing and information technology have developed rapidly. The unprecedented popularity and success of deep learning have established artificial intelligence (AI) as the frontier field of human exploration of machine intelligence. Meanwhile, thanks to the revolutionary progress of devices and the development of artificial intelligence (AI), brain-computer interface (BCI) implantation technology has also been rapidly implemented, which marks the beginning of the integration of BCI and AI, carbon-based and silicon-based. However, there are fundamental differences between the underlying logic of silicon-based and carbon-based computing, and the intelligent mechanism of the brain remains to be further explored. The visual cognition-guided twin AI deep network proposed in this study is a deep network technology driven by personal consciousness. It captures and analyzes individual thinking patterns and creative inspiration to create a unique visual world tailored for each user. In such an environment, everyone becomes the visual leader of their own created world, breaking the barriers between matter and consciousness, and expressing rich individuality and creativity.

The era of intelligent war is coming with the wide application of big data technology and artificial intelligence technology in military field. Based on the characteristics of intelligent war, this paper describes the challenges of traditional weapon equipment, and puts forward the development strategy of weapon equipment system under condition of intelligent war, it has certain practical significance to improve the system efficiency of weapon equipment and win the initiative of future wars.

Based on the formal description of the operational concept capability requirement analysis, a method of operational concept capability requirement analysis based on DRL(deep reinforcement learning) is designed. The key technologies of this method, such as simulation experiment, surrogate model, reinforcement learning, are analyzed and studied. Through the implementation of key technologies,small sample data sets with high reliability can be obtained through simulation experiments; Based on the experience data, the surrogate model of operation concept is constructed, and the model is optimized and trained by using multi-objective optimization algorithm with the high credibility simulation data set as the input; Finally, the surrogate model obtained from the training and the DRL framework are interactively optimized to achieve the reverse exploration of the operational concept capability requirements.

To solve the potential issue of reloading during mortar firing and ensure the safety of personnel and equipment during live firing in the military, a new type of mortar anti reloading system is designed. An indicator automatically switches its own state with the help of the sliding force during shell loading and buffering force during the movement of the buffering machine,undamentally solving the problem of "reloadingammunition" that may occur during mortar live firing. Through the interaction between the top rod component and the buffer machine, the buffering force can be transmitted to the anti reloading system. Through theoretical calculations and simulation experiments, it has been verified that its top rod component can meet the strength and stability requirements of mortar firing.

Due to its excellent concealment and difficulty in deployment and removal, mines have become a recognized challenge for the world navy in MCM operation. Airborne blue-green Lidar technology is one of the new technologies in ocean exploration. Due to its smaller impact on shallow water environments compared to traditional mine detection payloads, it can be used as a new type of mine detection payload to compensate for the shortcomings of traditional detection methods in shallow water areas. With the development of unmanned equipment, unmanned aerial vehicles equipped with airborne blue-green Lidar for mine detection will become one of the new mine detection methods in MCM operation. By analyzing the current development status of airborne blue-green Lidar in MCM operation at home and abroad, combined with the detection and positioning principles of airborne blue-green Lidar, this paper summarizes the application characteristics and combat effectiveness of unmanned aerial vehicles carrying blue-green Lidar in MCM operation. Finally, the development trend of future airborne blue-green Lidar technology is analyzed and forecasted based on the requirements of MCM operation.

Submarine-borne laser weapons can improve defense capabilities and increase the types of mission that can be completed. By introducing the research progress of laser weapons, the development status and application situation of laser weapons in the US, submarine installation requirements is proposed. The paper suggests timing for weapon use and analyzes the ability to damage anti-submarine aircraft. And it analyzes the ability to damage anti-submarine aircraft, and Provides reference for relevant research.

The intending goal of modeling and simulation is to create a unified Live-Virtual-Constructive (LVC) integration architecture, support rapid integration model and carry out simulation for joint flight training, tactical coordination, operational planning and evaluation. By exploring the connotation and application field of LVC, this paper analyzes the development plan of LVC training network, combs three typical application examples of LVC system, which are distributed mission combat, joint simulation environment and US Navy simulation training. Using the experience of LVC to solve the joint training, this paper gives the enlightenment of accelerating the construction of the joint training of our army, and provides valuable reference for the construction of our army's joint training environment.

Aiming at the situation of random changes in tactics and uncertain information in the confrontation pattern between missile and target in the air combat environment. By analyzing the motion relationship of a single missile attack to determine the target, the principles of dynamic game and differential game are introduced into the relative motion relationship between missile and target. In the dynamic game process of attack and defense confrontation, the terminal guidance problem of a single missile attacking and intercepting the target is modeled as a "one to one missile target" game model. The missile guidance law based on game theory under mixed strategy and the global strategic situation of bilateral optimization of the opponent at a certain moment are proposed. The method of model rolling prediction combined with differential game theory is introduced to simulate the confrontation between missile and target in uncertain attack and defense environments. The simulation results show that using this method can reduce the miss distance of missile to target and improve the hit accuracy of missile. The model provides a basis for missile attack and defense operations.

With the widespread application of cruise missile technology and the increasing complexity of mission execution, the mission planning of cruise missile formations is facing new challenges. A multi patrol missile pre war planning task allocation model based on multiple constraints is constructed, with the optimization objectives of mission value return and mission loss cost, while considering various practical constraints such as task execution timing. To solve the model, an improved genetic algorithm based on simulated annealing is designed, which adjusts the adaptive selection strategy and adaptive crossover strategy based on the population fitness of different periods, ensuring convergence while increasing population diversity. Finally, the effectiveness of the designed improved algorithm in solving the task allocation problem of multiple patrol missiles was verified through a case study.

Aiming at the characteristics of large solution space, discrete, dynamic and nonlinear of firepower-target assignment problem, this paper proposes a deep reinforcement learning algorithm based on DQN. By combining the 6-layer fully connected feedforward neural network with the Q-learning algorithm, the perception ability of deep learning and the decision-making ability of reinforcement learning are fully utilized. Through the comparison of model performance tests, this method has strong fitting ability, fast convergence speed and small variance jitter, and the distribution results meet the combat expectations, which can provide some reference for commanders to make decisions on fire strike problems.

In response to the problem of limited evaluation indicators for the cost function of a single route planning algorithm, and the recommended optimal route for planning being the optimal local indicator, a land attack cruise missile route planning evaluation index system is established based on missile operational requirements and operational environment characteristics, starting from flight safety, penetration safety, and strike accuracy requirements. Qualitative analysis is conducted on the probability of missile hitting the ground and falling into the human zone, as well as the amplitude of trajectory crossing the detection and interception zones specific evaluation indicators such as autonomous flight and detection recognition accuracy have certain reference significance for optimizing route planning and optimization algorithms.

In modern military warfare, the pattern of air-ground coordination with multi-formation has become more and more important. However, the existing target intention recognition methods are effective for single formation, but lack of effective solutions for multi-formation scenarios with air and ground coordination. In this paper, Dynamic Series Bayesian Network is used to identify the intention of air-to-ground cooperative formation. This method firstly constructs an overall model of intention recognition of air-to-ground cooperative formation by using DSBN, which is used to describe the cooperative action process between air and ground formations. Then, events in different battlefield domains and their related probability relations are fused together with auxiliary battlefield information. The inference network is used to recognize the intention of enemy cooperative formation. This method fully considers the behavior rules of the air target, describes its behavior pattern and trend in detail, and is more suitable for the scenario of multi-cooperative target formation. Finally, the feasibility and effectiveness of this method is verified by simulation example.

Target maneuver type recognition is a key aspect in the fields such as object tracking and intention recognition. A multi-layer label generation method based on improved K-medoids clustering is proposed to address the problems of insufficient universality and low accuracy of existing maneuver type recognition methods. Taking the maneuver trajectory of aerial targets as an example, various data preprocessing methods are designed, and dynamic time warping is used as the distance metric for clustering algorithm. By constructing a label framework to guide the implementation process of the clustering algorithm, iterative clustering is carried out to generate multi-layer labels. The algorithm is tested on a public dataset, and the experimental results showed that the recognition accuracy of the first layer label achieved 89.75% under unsupervised conditions, which is closed to traditional supervised algorithms; Meanwhile, compared to the condition without introducing a label framework, it is more effective in distinguishing the second layer of fuzzy labels.This method requires only a small amount of expert knowledge and can be easily extended to different fields and different maneuver type.

Two bearing-only maneuver detection methods based on deep learning are proposed to address the problems of long detection delay and low accuracy of existing bearing-only maneuver detection methods for underwater targets. The bearing observations of the target in the constant velocity (CV) motion state and constant turning (CT) motion state are used as the training data set. The target motion pattern classification and bearing prediction are realized through the Long short-term memory (LSTM) neural network, and then realize the maneuver detection of underwater targets based on motion pattern classification and bearing prediction. The simulation results show that compared with the traditional bearing prediction maneuver detection method, this method reduces the bearing observation error and has a lower sensitivity of target maneuver magnitude, and has a higher maneuver detection accuracy and reduces the maneuver detection delay.

Aiming at the lack of effective analysis and identification of operational area in the traditional battlefield situation awareness process, a method of aerial battlefield target trajectory analysis method for operational area extraction is proposed. Firstly, based on the improved Douglas-Peucker algorithm, the featured trajectory is extracted based on the original target trajectory, thus reducing the trajectory redundant information. Secondly, the clustering by fast search and find of density peaks (CFSFDP) algorithm is used to cluster the featured trajectory, which obtains multiple featured points clusters. Finally, each featured points cluster is scanned based on Graham's convex hull algorithm to obtain closed polygons to characterize the combat areas. The feasibility and effectiveness of the algorithm model are verified by several case simulations.

In order to solve the target tracking problem of multiple formations in the context of unclear nascent target parameters, a formation target tracking algorithm with unknown nascent target-related parameters is proposed, which can effectively improve the tracking performance and thus better meet the demand. By using the Gaussian mixed probability hypothesis density (ET-GM-PHD) filtering framework, the PHD parameters are predicted and then the correlation between each Gaussian component is analyzed to filter out the optimal set of measurements and finally eliminate most of the noise disturbances. Based on the characteristic of dense distribution measured in the movement area of the formation target, the location of the formation target is found and tracked by the target detection method. Finally, the simulation experiments show that the method in this paper can track the formation target quickly and effectively with good tracking performance when the parameters related to the nascent target are unknown.

The main challenge of multi-object tracking (MOT) is identity switch caused by severe occlusion. The solution to identity switching is video object association, which assigns an identity number to the same target in different frames. In this paper, a long-short time association algorithm is proposed for identity switching. In the short-time, that is, the motion features between adjacent frames are used to match, and in the long-time, that is, the non-adjacent frames are directly added to the appearance features for association to rematch the object detected after occlusion. Besides, the Kalman filter is improved and the frame width parameter is added to make the predicted frame more accurate; appearance features use average appearance features and increase detection confidence as update parameters to make appearance more robust and can still work in complex scenes. The new tracker, LSATrack, achieves 81.3MOTA and 81.3IDF1 in the MOT17 and achieves stable tracking in severe occlusion scenarios.

Addressing the issue of relatively low recognition rates for named entities in domain-specific short texts under resource-constrained computational environments, a novel hybrid model combining a Dual BiLSTM_CRF architecture with a fully connected network is designed to identify named entities in these texts. The model leverages critical knowledge entities and their key relationships from a domain knowledge graph, which undergoes projection transformation, clustering, and global vector word embedding processing. Based on the calculation of word vector similarities, it identifies similar critical knowledge entities to those being recognized within the domain. These identified knowledge entities are then substituted into the original domain short text, generating an augmented version that is fed, along with the unmodified text, into the model for named entity recognition. This integration of domain knowledge into the recognition process of named entities in domain-specific short texts has shown promising results. Experimental outcomes demonstrate that this method outperforms existing comparable approaches in terms of its enhanced identification capabilities.

Aiming at the current situation of lack of environment and equipment in radar training, a radar operation training method based on the combination of virtual and real is proposed. Firstly, the connotation of the concept of virtual-real combination is analyzed. Secondly, the interaction methods of combining physical radar and virtual environment and virtual radar and physical environment are studied. Finally, combined with the specific training requirements, the application of real-real, real-virtual, virtual-real, and virtual-virtual radar interaction methods in radar training is analyzed in the same scenario.

In order to improve the recognition accuracy of radar targets and the recognition performance under incomplete angular domain data, a high resolution range profile (HRRP) recognition algorithm based on time series forest is proposed. The basic principle of time series forest algorithm and the basic steps of HRRP target recognition are introduced in detail. The experimental results of measured HRRP data show that compared with k-nearest neighbors(KNN), support vector machine(SVM), random forest(RF), convolutional neural network(CNN), long short-term memory(LSTM) and other classical target recognition algorithms, the proposed algorithm has better recognition performance and better angular domain promotion ability. Under the condition of training only 1/2 and 1/3 angular domain data, the average recognition rate of the full angular domain data is better than 85%, which is an average increase of 5.2% compared with the above method.

The network topology is marked with the device type, which is difficult to reflect the service relationship between the device nodes. Aiming at this problem, this paper proposes a network service node division algorithm based on service mapping. The algorithm judges the ownership of the service community through the service mapping relationship between network nodes, and finds the key nodes of network services and the subordinate relationship between nodes. This paper introduces virtual community based on traditional membership matrix. The network service community membership influence degree algorithm not only give the community membership degree of nodes, but also reflect the global influence degree of nodes, providing a basis for the selection of key network nodes in network security. In the experiment based on LFR network, this paper compares the algorithm sensitivity and performance with similar algorithms, and validates the effectiveness of algorithm partitioning in the Internet2 OS3E simulation experiment by simulating denial of service attacks. The comparative experimental results show that the service mapping algorithm has the characteristics of low sensitivity and high practicability in threshold selection, and its performance is better than that of similar algorithms. The simulation results show that the service community partition is effective and can give the correlation between node services.

Geosynchronous orbit satellite is usually the first early warning equipment to catch ballistic missile launches in actual combat. However, the detection information it provides is lack of dimension, which makes the traditional target guidance method hard to be directly applied on the target guidance for low-orbit satellite. Aiming at this point, this paper studies the target guidance method for low-orbit satellite by geosynchronous orbit satellite. First, based on the altitude constraint of infrared radiation during the launch and active flight of the ballistic missile, the azimuth and pitch detection ranges of the infrared sensor of the low-orbit satellite are calculated with the guidance of one geosynchronous orbit satellite. Then, the detection distance of the infrared sensor of the low-orbit satellite is modeled and calculated, which leads to a successful target detection for the low-orbit satellite. Finally, the effectiveness and accuracy of this method are verified through a numerical simulation experiment.