International Journal of Engineering Intelligent Systems

AgriSpot-Trans: Fine-Grained Micro-Lesion Segmentation in Complex Unstructured Agricultural Environments via Occlusion-Aware Hierarchical Transformer

Xianglin Qu — 2026-05-01

Accurate pixel-level segmentation of early-stage micro-lesions in unstructured agricultural environments is critical for precision agriculture, yet it remains fundamentally challenged by severe field occlusions, optical interferences, and fuzzy biological boundaries. To address these issues, we proposeAgriSpot-Trans, a novel end-to-end occlusion-aware hierarchicalTransformer. Unlike conventional hard-cropping pipelines that disrupt global canopy topology, our architecture introduces an Environment-AwareToken Routing (E-Decoupler) module. Driven by a Gumbel-Softmax mechanism, it softly suppresses background noise and specular reflections in the feature space. Furthermore, to overcome the morphological mismatch of standard
square-window attention, we design Morphological Radial Attention (MRA), which dynamically aligns its star-shaped receptive field with the radial diffusion gradients characteristic of biological tissue damage. Finally, a Level-Set Signed Distance Function (SDF) Decoder, optimized by a Boundary Aware Active Contour loss, is employed to achieve continuous, sub-pixel contour evolution. Extensive experiments on PlantVillage, RoCoLe, and our newly curated Complex-Field-Pest (CFP) benchmark demonstrate that AgriSpot-Trans achieves a state-of-the-art mIoU of 78.2% in complex field scenarios—a 9.5% absolute improvement over the strong SegFormer baseline (p < 0.01). Most notably, it boosts the recall for extreme micro-targets
(= 5 px) by an unprecedented 23.7%, delivering a highly robust and edge-efficient visual perception framework for real-time, variable-rate pesticide application.

Keywords: Precision agriculture, phytopathology, micro-lesion segmentation, vision transformer, level-set method, unstructured environments, active contour.

Cite As

X. Qu, H. Peng, X. Zhu, X. Shen, X. Chen, J. Luo, R. Zhou, "AgriSpot-Trans: Fine-Grained Micro-Lesion
Segmentation in Complex Unstructured Agricultural Environments via Occlusion-Aware Hierarchical
Transformer", Engineering Intelligent Systems, vol. 34 no. 3, pp. 289-303, 2026.

Automated Generation and Optimization Methods for Smart Contract Vulnerability Code Integrating Large Models and RAG Framework

Zuohu Chen — 2026-05-01

The application of big models to vulnerability generation for smart contracts is hampered by a lack of knowledge within the field, regularly producing difficulties such as logical anomalies, conceptual misunderstandings, and poor concealment within the generation results. In order to overcome the discrepancy between the poor quality of generated results and the important role played by big models within the field, the article proposes a retrieval-enhanced generation optimization strategy that combines a hybrid retrieval strategy and a strategy for a prompt and programmatic validation cycle constrained by pre-defined generation objectives. The results demonstrate that the strategy effectively enhances the authenticity level from 0.342 for the benchmark model CodeLlama (Code Large Language Model Meta AI)-13B to 0.867 and enhances the concealment score from 0.106 to
0.289, which contributes to improving the recognition capability level within the field by improving the quality and concealment level of generation, supplying a useful model for the proper use of big models within a relevant field. This research work presents a systematic optimization strategy for automatically optimizing the generation of data concerning smart contract security tests and adversarial examples. In addition to optimizing generation quality and concealment level, it functions as a relevant model for big models within a relevant field.

Keywords: Smart contract security; vulnerability code generation; large language model; search enhanced generation; automated testing

Cite As

Z. Chen, Z. Peng, Y. Dou, N. He, W. Li, "Automated Generation and Optimization Methods for Smart Contract Vulnerability Code Integrating Large Models and RAG Framework", Engineering Intelligent Systems, vol. 34 no. 3, pp. 305-318, 2026.

Construction of Computer Vision Security Model Based on Machine Learning and Artificial Intelligence Networks

Nan Si — 2026-05-01

Computer vision is being widely used today, but due to the inadequacies of computer vision systems, security issues have become a serious challenge. In this current study, a computer vision security model based on machine learning and artificial intelligence (AI) networks was constructed to improve the security and robustness of existing computer vision models, and provide effective solutions for computer vision problems in practical application scenarios. Moreover, the technology application of machine learning and AI network in the construction of computer vision security model was analysed to build a computer vision security model with four parts: data acquisition, data pre-processing, network construction and model training. Three real-life cases were selected for experimental analysis to verify the advantages of the proposed model. The accuracy, precision and recall of
100 facial samples using a computer vision security model constructed using machine learning and AI network technology were compared with those using traditional computer vision security models. According to the results, the accuracy, precision and recall of the proposed computer vision security model improved by 18%, 16% and about 18.2%, respectively. These findings indicate that the computer vision security model based on machine learning and AI networks constructed in this paper offers significant advantages in terms of object detection and good performance. The proposed model is of great relevance for the current security issues facing computer vision systems, providing new ideas and methods for research and practice in computer vision security.

Keywords: machine learning, artificial intelligence network, face recognition, computer vision security model, model construction

Cite As

N. Si, L. Tan, L. Yin, Z. Liu, "Construction of Computer Vision Security Model Based on Machine Learning and Artificial Intelligence Networks", Engineering Intelligent Systems, vol. 34 no. 3, pp. 319-329, 2026.

Construction of Humanoid Robot and Generative Artificial Intelligence Information System Based on Deep Learning

Huijuan Sun — 2026-05-01

With the rapid development of DL technology, artificial intelligence (AI) and robotics are gradually merging, especially the combination of humanoid robots and generative AI, which is becoming a key technology to promote the intelligent transformation of future industries. However, in traditional research, humanoid robots and generative AI technologies are developed in isolation, lacking effective collaboration and natural interaction, and lacking innovation in cross-domain application. To address these problems, in this study, an information system is constructed based on DL humanoid robots and generative AI. The system consists of three modules: a perception, decision and action module, and generation and interaction modules.
The perception module collects and preprocesses data of multiple modalities and then fuses the multimodal data. The decision and action module achieves the decision-making function of the information system through the DQN model. The generation and interaction module introduces the Generative Pre-Trained Transformer (GPT) model to realize the generation and interaction functions of the information system. Experiments have shown that the image perception ability of the system constructed in this study is good. The average perception error of 100 areas in the experiment is 0.0361. The decision-making ability is strong, and the decision-making score for different scenarios exceeds 0.83. It can also maintain good consistency in the generated content while taking emotions into account during conversations. The robot equipped with the system proposed in this paper and the traditional robot were put into the service industry for comparative experiments. The results demonstrated that the robot equipped with the proposed system is superior to the traditional robot according to all scoring indicators. The information system of humanoid robots and generative AI based on DL constructed in this study breaks through the limitations of the isolated development of traditional humanoid robots and generative AI, through the strong integration of deep perception, decision-making and interaction. It significantly improves the intelligence level of robots, hasbroad application prospects, and provides new ideas and technical support for promoting industrial intelligent transformation and cross-domain application innovation.

Keywords: Deep learning, Humanoid robot, Generative Artificial Intelligence, Multimodal Perception, Deep Q network, Generative Pre-Trained transformer

Cite As

H. Sun, J. Shen, "Construction of Humanoid Robot and Generative Artificial Intelligence Information System Based on Deep Learning", Engineering Intelligent Systems, vol. 34 no. 3, pp. 331-340, 2026.

Deep Learning-Based Architectural Visual Style Transfer Simulation and Functional Performance Optimization

Jing Zhao — 2026-05-01

Architectural style transfer methods have difficulty achieving a balanced optimization of both visual aesthetics and functional performance, resulting in frequently unpragmatic solutions that fail to meet real-life architectural needs. The paper explores the unique difficulties experienced by researchers engaged in architectural visual style transfer simulation and proposes a multi-objective optimization framework utilizing deep learning that addresses these difficulties. This paper addresses the issue of maintaining a high level of fidelity while adhering to physical constraints without jeopardizing structural safety. The proposed framework utilizes a CycleGAN (Cycle Generative Adversarial Network), which incorporates an attention mechanism
to increase the sensitivity to the transfer of local characteristics, and combines matrix representations with a VGG-19 (Visual Geometry Group 19) network to provide two types of loss (style and content) that ensure a high level of visual fidelity. The NSGA-II (Non-Dominated Sorting Genetic Algorithm II) algorithm is included in the framework to dynamically optimize both architectural style and functional performance by using energy use anddaylighting calculated by EnergyPlus and Radiance as hard constraints. Achieving this will require aclosed-loop generation-simulation-evaluation feedback process. Real-time imports of style transfer designs go into a Building Information Modeling (BIM) andVirtual Reality (VR) simulation where EnergyPlus and Radiance will compute the designs’ performance. The performance data are then used as feedback for the NSGA-II algorithm to
modify generator settings and produce an optimal design. Finally, the architectural model shows the trade-offs between the aesthetics, energy use, and daylight use of the various design solutions to provide designers with visual assistance for deciding upon which is best. Experimental results indicate that this approach resulted in better scores in user satisfaction surveys and 16.4% less EUI than pre-transfer models. The post-transfer sDA increased 8.2% as compared to StyleGAN and average use of space increased 22.3% over pre-transfer models. The average maximum deformation was 7.8 mm and average maximum stress was 18.1 MPa, with no detrimental effect to the safety of the building structure. The research provides both theoretical
and practical tools supporting intelligent building design that offers both aesthetic and practical value.

Keywords: Deep learning; style transfer; architectural simulation; architectural functional performance optimization; generative adversarial network

Cite As

J. Zhao, "Deep Learning-Based Architectural Visual Style Transfer Simulation and Functional Performance Optimization",
Engineering Intelligent Systems, vol. 34 no. 3, pp. 341-352, 2026.

Design of Resource Intelligent Systems Recommendation Algorithm and System Based on Heterogeneous Network Representation Learning

Jiana Bi — 2026-05-01

Based on the historical behavior data ofusers, the recommendation systemfilters outvaluable information from complex data andactively recommends resources for users, so that they can obtain resources quickly and accurately. This effectively solves the problem of information overload in the Internet era, and improves the efficiency of resource use. A heterogeneous information network can naturally model the complex objects and their rich relationships in the recommendation system, and explore the potential connections between users by analyzing different types of relationship paths.
This paper focuses on two issues: Firstly, the recommendation algorithm is studied, including recommendation algorithm design, model training design, experimental data set selection, evaluation index design, and the superiority of the algorithm is verified through experiments. Then we study the recommendation system, design the simulation function which consists of the recommendation use subsystem and the recommendation management subsystem, and verify the validity of the software through the system test.

Keywords: heterogeneous network representation learning; resource recommendation algorithm; system design; random walk; Skip-Gram model; experiment and testing

Cite As

J. Bi, "Design of Resource Intelligent Systems Recommendation Algorithm and System Based on Heterogeneous Network Representation Learning", Engineering Intelligent Systems, vol. 34 no. 3, pp. 353-359, 2026.

Identification of Built-up Urban Areas in High-Resolution, Remotely- Sensed Images Based on Deep Learning

Chujun Li — 2026-05-01

Theaccuracy of traditional identification methods for built-up urban areas is often challenged when faced with complex urban environments comprising dense buildings and shadows, especially when identifying small targets and multi-scale building features. To address these problems, this paper constructs an built-up urban area identification model based on the Weighted Feature Fusion Enhanced UNet (WFFE-UNet). The data used in this paper isfrom the Massachusetts Buildings Dataset, whichprovides aerial images oftheBoston metropolitan areawith aresolution of1 meter. Histogram equalization and principal component analysis denoising techniques are used to improve image quality and help the model capture key information
in the image more accurately. Then, an identification model based on WFFE-UNet is constructed. Based on the UNet framework, this model uses an improved convolution module with multi-scale perception to enhance the ability to express multi-scale building structures in high-resolution images. The Convolutional Block Attention Module (CBAM) attention mechanism is integrated to improve the identification ability of small targets. By combining channel attention and spatial attention, the model’s sensitivity to the key features of small targets is significantly enhanced. Finally, the strategy of joint optimization of cross-entropy loss and Dice loss is adopted to improve the model’s convergence efficiency and identification precision. Experimental results show that the overall accuracy, precision, recall, and intersection over union (IoU) of the WFFE-UNet model are 97.34%, 96.76%,
95.42%, and 82.65%, respectively. In the cross-domain test, the overall identification accuracy of the model in different cities remains above 96%. For small targets B and P, the IoU of the model studied in this paper is 78.98% and 78.46%, respectively. The built-up urban area identification method based on WFFE-UNet and CBAM attention mechanism successfully solves the problems encountered by traditional methods when applied complex scenes by combining automatic feature learning, multi-scale information fusion, and an attention mechanism.

Keywords: deep learning, high-resolution remotely-sensed image, built-up urban area identification, Weighted Feature Fusion Enhanced UNet, Convolutional Block Attention Module

Cite As

C. Li, M. Gapar, M. Johar, "Identification of Built-up Urban Areas in High-Resolution, Remotely- Sensed Images Based on Deep Learning", Engineering Intelligent Systems, vol. 34 no. 3, pp. 361-374, 2026.

Low-Altitude Status Monitoring and Fault Diagnosis Algorithm for Aircraft Equipment Based on Big Data Mining and Analysis

Wenjing Yin — 2026-05-01

Themonitoring and fault diagnosis of aircraft equipment at low-altitudes refers to the monitoring of the relevant parameters of various operating states of aircraft equipment, effectively determining its operational status based on existing data, and the diagnosis of any faults, so that appropriate technical remedial measures can be taken. This study examines the analysis and application of aircraft equipment low-altitude state monitoring and fault diagnosis methods based on big data mining (DM), and establishes relevant systems. The parameters of aircraft equipment were analyzed by applying association rules, whereby the rules governing the operation of equipment were found. The verification analysis and experimental data of the system
showed that in the test set, when the number of aircraft equipment fault samples was 200, the correct number of fault samples for the traditional system and the system proposed in this paper was 171 and 191 respectively. The fault diagnosis time of the two systems was 1.127s and 0.562s respectively. This indicates that the monitoring and fault diagnosis based on big DM analysis of aircraft equipment in the low-altitude condition has certain feasibility and merits further promotion and application.

Keywords: aircraft equipment, big data mining analysis, fault diagnosis algorithm, low altitude condition monitoring

Cite As

W. Yin, S. Lv, X. Li, "Low-Altitude Status Monitoring and Fault Diagnosis Algorithm for Aircraft Equipment Based on Big Data Mining and Analysis", Engineering Intelligent Systems, vol. 34 no. 3, pp. 375-384, 2026.

Modeling of Integrated Energy Distribution Network and Power System Security and Stability Control Based on Perception System

Gang Wang — 2026-05-01

In response to the problems of lagging data collection and monitoring, lack of dynamism in system models, and long fault handling and recovery time in traditional integrated energy distribution network modeling, this study constructed a perception system-based integrated energy distribution network model and power system safety and stability control method. Firstly, by deploying multiple sensors at key nodes in the distribution network, real-time monitoring of power load, thermal energy flow, and gas transmission was achieved. The collected data was then fused using the Kalman filter and data fusion algorithms (such as Bayesian fusion) to eliminate noise and redundant information, resulting in high-quality system state estimation.
Then, the DreamerV2 algorithm was used to establish a dynamic model of the integrated energy distribution network, which simulated the energy flow and control strategies in the environment for long-term system optimization prediction. Finally, based on the data results after the DreamerV2 model prediction, the MILP (Mixed-integer Linear Planning) method was applied for optimization. Experimental results showed that the dynamic model had good processing capacity, high system stability. The transition response time was only 2.5 seconds, energy efficiency was about 85.5%, MTTR (Mean Time To Repair) was 1.69 hours, the average load tracking accuracy was 1.89%, the data proved the method in improving the security and stability of the power system.

Keywords: Electric Power System; Distribution Network Modeling; Deep Reinforcement Learning Algorithm; DreamerV2 Model; Data Acquisition

Cite As

G. Wang, X. Wang, Z. Chen, Y. Li, Z. Peng, "Modeling of Integrated Energy Distribution Network and Power System Security and Stability Control Based on Perception System", Engineering Intelligent Systems, vol. 34 no. 3, pp. 385-297, 2026.

Power Indicator Benchmarking Evaluation System Based on Multi-Source Data Fusion

Xiaomeng Liu — 2026-05-01

Traditional power indicator benchmarking systems are unableto meet the multi-dimensional dynamic evaluation needs of new power systems. In this paper, a dynamic benchmarking evaluation system is constructed based on multi-source data fusion. By establishing a multi-dimensional indicator library of “source-grid-load-storage-carbon”, integrating data from several domains such as meteorology, geography, and policy, an integrated methodological framework comprising dynamic weighting, comprehensive evaluation, and trend prediction is designed to achieve adaptive assessment and the analysis of future system status. Empirical results show that this system can improve the coverage of traditional evaluation methods for dimensions such as clean and low-carbon energy from 5 points to over 75 points, and effectively support risk warning and proactive dispatch by
means of ongoing prediction of key indicators such as the rate of utilization of renewable energy (RMSE=0.18%).

Keywords: Data Fusion; Power System Evaluation; LSTM Prediction; Dynamic Benchmarking Framework; Renewable Energy Utilization

Cite As

X. Liu, J. Zhang, X. Kou, Z. Peng, T. Liu, "Power Indicator Benchmarking Evaluation System Based on Multi-Source Data Fusion",
Engineering Intelligent Systems, vol. 34 no. 3, pp. 399-411, 2026.

Research on Fault Warning System for Hydroelectric Facilities Based on Cloud Computing and GAN

Qiuyue Zhong — 2026-05-01

Because hydropower facilities are one of the country’s important energy infrastructures, the safety of their operations is crucial to social and economic development, the security of lives and property, and the protection of the ecological environment. Traditional safety management methods usually rely on manual monitoring and a single system, resulting in long response times, difficulty in information integration, and difficulty in timely capturing potential risks. This approach often has information islands and data fragmentation problems, and cannot achieve comprehensive, real-time monitoring and early warning of facilities, increasing the probability of safety hazards. This study designed and implemented a hydropower facility safety management platform based on cloud computing. A distributed storage structure was used to store collected data and separately build a Kafka transmission architecture to connect the various departments of the platform and speed up the information transmission within the platform. It used generative adversarial network (GAN) technology to identify faults and issue early warnings through the transmission architecture. The experimental results show that during the data transmission process, the average transmission accuracy is 94.6%, the average transmission completeness is 98.2%, and the average transmission synchronization is 97.8%. At the same time, when the platform load is low, the response time can be as short as 0.89 seconds. The evaluation of the platform’s early warning capability showed that the AUC value of fault identification is 0.92. This study analyzed key indicators such as throughput, latency, and CPU usage of the platform, and demonstrated that the proposed system can still maintain stable operation in high load and complex environments. Through the application of cloud computing and distributed technology, the proposed hydropower facility
safety management platform significantly improves warning capabilities and system stability, ensuring efficient operation of the platform in complex environments and providing intelligent and real-time solutions for the safety management of hydropower facilities

Keywords: Hydropower facility; cloud computing; security management platform; generative adversarial networks; early warning capability

Cite As

Q. Zhong, J. Zhao, "Research on Fault Warning System for Hydroelectric Facilities Based on Cloud Computing and GAN",
Engineering Intelligent Systems, vol. 34 no. 3, pp. 413-422, 2026.

Utilizing Intelligent Vector Image Technology to Graphically Display Network Monitoring

Yao Zhang — 2026-05-01

Traditional network monitoring graphic display methods have the problems of poor display effect, difficulty in manual identification and tracking of targets, and poor real-time performance. In this study, a network monitoring graphic display system was designed based on intelligent vector image technology to solve the problems in traditional image display methods. An adaptive histogram equalization algorithm was used to denoise and sharpen an image, and remove image distortion to improve the graphic display effect. Then, the Faster R-CNN target detection model based on convolutional neural network was used to achieve automatic target detection and tracking. The real-time image was processed through an image processing pipeline, and finally the target object in the monitoring image was converted into a vector graphic using intelligent vector image technology.
Through experiments, the network monitoring graphics display system designed in this study was compared with traditional monitoring systems and raster image display systems in terms of performance, stability, image compression rate, and user experience. In terms of performance testing, the proposed system outperformed traditional monitoring systems in terms of image quality, rendering efficiency, real-time performance, loading speed, and display delay. In regard to stability, the proposed system performed the best in terms of data integrity, reaching 96.7%, which was significantly higher than the other two types of systems, with a failure rate of only 4.3%. Regarding system image compression rate, the average compression coefficient of the proposed system was 89.7%, which was higher than 56% of traditional monitoring systems and 72.7% of raster image display
systems. In terms of user satisfaction scores, the proposed system performed best. The experimental results showed that intelligent vector image technology can significantly improve the effectiveness and performance of network monitoring graphics display, offering new ideas and methods for the improvement of network monitoring systems and has broad application prospects.

Keywords: Intelligent Vector Image Technology; Network Monitoring; Graphic Display; Feature extraction; Image Quality

Cite As

Y. Zhang, "Utilizing Intelligent Vector Image Technology to Graphically Display Network Monitoring",
Engineering Intelligent Systems, vol. 34 no. 3, pp. 423-435, 2026.