A method of identifying and positioning coal and gangue on mobile devices is provided to address the difficulties of high complexity and difficult deployment of existing machine vision algorithms. Combining CSPDarknet53 and the GhostNet embedded with efficient channel domain attention to build a lightweight feature extraction network and using the Meta-ACON activation function to adapt the network’s nonlinearity at each layer. The detection head’s expression capability is enhanced using a decoupled head. The actual position coordinate relationship equation of coal and gangue is deduced by borrowing the center point object box positioning principle. Build an experimental setup, collect and construct coal and gangue datasets, and use Mixup data augmentation to improve the network’s ability to position coal and gangue for stacking cases. The models are trained, and tested for different lighting conditions, sizes, and stacking cases, then deployed on the mobile device. The results show that the model has the highest detection accuracy and can correctly identify and position both coal and gangue under complex conditions with a high confidence value. The FLOPs is 76.83% lower compared to the original network YOLOv5s. The mAP reaches 0.996 and the FPS is 40.11. The inference time on the mobile device is reduced to 228 ms, which basically meets the speed requirement of coal gangue identification and positioning.

The intelligentization of coal mines is the only approach to the high-quality development of the coal industry. Detection, identification, and sorting of coal gangue is an important part of the intelligentisation of coal mines. For the problems of traditional target detection algorithm model, such as high complexity, difficult training, and poor identification and positioning effect on coal and gangue under complex conditions, an identification and positioning method for coal gangue based on lightweight mixed domain attention was proposed. First, a mixed domain attention model was constructed through residual connection of spatial attention and efficient channel attention, and the MobileNetv3 network was optimized to improve its detection effect on coal and gangue under different conditions. Secondly, the optimized network was used to replace the feature extraction network in YOLOv4, and lightweighted using depthwise separable convolution to reduce model complexity and thereby obtain the ECSMv3_YOLOv4. Then the relationship between the actual position of coal and gangue on the conveyor belt was derived. Finally, after an experimental device was set up, a multi-condition coal and gangue dataset was built, the model was trained, as well as the identification and positioning results of the model under different sizes and light intensities were tested, and compared with those of other algorithms. Experiment results showed that the algorithm has the highest detection accuracy and efficiency, and can correctly identify and position coal and gangue under complex conditions with high confidence values. Compared with other algorithms, the mAP value of the proposed method is increased by 2.84% at most. The FLOPs was only 2,610,046, and the FPS of detection reached 28.8. Accordingly, this research is considered to be of certain theoretical value and technical reference for identification and positioning of coal gangue.

Matching constraints in an analog integrated circuit (IC) are critical to optimizing layout performance. To extract these matching constraints accurately and efficiently from the netlist, we propose the heterogeneous matching constraint extraction graph neural network (MCE-HGCN). First, the netlist is mapped into a heterogeneous attribute multi-graph, and based on the characteristics of analog IC matching constraints, a mixed-domain attention mechanism is developed to leverage both the topology information and node attributes in the graph to characterize node embeddings. A matching classifier, implemented using the support vector machine (SVM), is then employed to classify different types of matching constraints from the netlist. Additionally, a matching filter is introduced to remove interference terms. Experimental results demonstrate that the MCE-HGCN model converges effectively with small datasets. In the matching prediction process, the mean F1 score reached 0.917 across different netlist processes and circuit types while maintaining a shorter runtime compared to other methods. Ablation experiments also show that incorporating the mixed-domain attention mechanism and the matching filter individually leads to significant performance improvements. Overall, MCE-HGCN excels at extracting matching constraints from various analog circuits and processes, offering valuable insights for placement guidance and enhancing the efficiency of analog IC layout design.

In the design of analog integrated circuits, the placement stage in layout design plays a crucial role in determining the final circuit performance. Current automated placement algorithms are primarily driven by geometric constraints, often requiring repetitive iterations to optimize circuit performance. Accelerating the design process necessitates accurate evaluation of placement quality; however, the complex relationship between circuit performance and layout makes rapid and precise assessment challenging. Traditional methods for evaluating placement quality require the completion of the entire layout design, involving time-consuming processes such as design rule check, layout versus schematic, parasitic parameter extraction, and simulation. The original graph isomorphism network (GIN) model's neglect of edge features and reliance on static aggregation mechanisms lead to incomplete topology-physics representations and degraded accuracy in predicting performance metrics. To address the issues, we introduce a modified graph isomorphism network (MGIN) that utilizes edge feature concatenation and adaptive weighted loss adjustment for multitask regression. MGIN effectively predicts multiple post-layout performance metrics using carefully crafted placement features. In comparison to baseline models, it achieves lower average relative error rates for operational transconductance amplifier circuits. Our results demonstrate an average root mean square error of 3.549 and a relative error rate of 3.549% across multiple performance metrics. Our model achieves superior prediction accuracy over GIN, with an 80.4% lower average error. By mapping layout configurations directly to post-layout performance, the proposed MGIN offers specific and actionable guidance for iterative redesign in analog integrated circuit development.