Aiming at the current problems of low detection accuracy of high-voltage cabinet switches and large models that are difficult to deploy, a high-voltage cabinet switch detection method based on the improved EfficientNet-YOLOv5s model is proposed. Firstly, offline data expansion and online Mosaic data enhancement are performed based on the high-voltage switchgear images obtained by video frame extraction to alleviate the problems of unbalanced sample categories and the small number of samples. Secondly, the backbone network of the YOLOv5s model is replaced with a lightweight EfficientNet-B0 network, and the original convolution layer is replaced by depthwise separable convolution, and the original Squeeze-and-Excitation Networks were replaced with Efficient Channel Attention to optimize the efficiency and accuracy of the network model. To enhance the positioning information of small objects without bringing too much computational burden, the proposed improved multi-scale feature fusion Neck network uses the Bi-directional Concatenation(BiC) module to aggregate the feature maps of three adjacent layers. Finally, the H-Swish function is used to replace the Swish function in the MBConv structure, avoiding the exponential operation of the sigmoid function, and further improving the calculation efficiency of the model.
Accurate and timely lane detection is imperative for the seamless operation of autonomous driving systems. In this study, leveraging the gradual variation of lane features within a defined range of width and length, we introduce an enhanced Spatial-Temporal Recurrent Neural Network (SCNN) framework. This framework serves as the cornerstone of an innovative hybrid spatial-temporal model for lane detection, which is tailored to address the prevalent issues of substandard detection performance and insufficient real-time processing in intricate scenarios, such as those involving lane erosion and inconsistent lighting conditions, which often challenge conventional models. With the foundational understanding that lanes manifest as continuous lines, we employ a temporal sequence of lane imagery as the input to our model, thereby ensuring a rich provision of feature information. The model adopts an encoder-decoder structure and integrates a Spatial-Temporal Recurrent Neural Network module for the extraction of interrelated information from the image sequence. The model culminates in the output of the lane detection results for the terminal frame. The proposed lane detection model exhibits a commendable synthesis of accuracy and real-time efficiency, attaining an Accuracy of 97.87%, an F 1 -score of 0.943, and a FPS of 19.342 on the tvtLANE dataset and an Accuracy of 98.21%, an F 1 -score of 0.957 on the Tusimple dataset. These metrics signify a superior performance over a majority of the current lane detection methods.
To address the challenges posed by complex backgrounds and target scale variations in remote sensing aircraft detection, we propose an improved YOLOv7-tiny network designed to boost both detection accuracy and the model’s ability to generalize across varying conditions. Complex backgrounds hinder effective target feature extraction, while variations in target scale complicate the real-time detection of objects at different sizes. To address these issues, we introduce a multi-head dynamic joint self-attention mechanism that combines self-attention, channel attention, and spatial attention, thereby enhancing the model’s ability to capture crucial target features. Furthermore, we design an enhanced feature extraction module that expands the receptive field and improves the model’s ability to extract features from multi-scale targets. In addition, we constructed diverse remote sensing aircraft detection datasets and applied a range of data augmentation techniques to improve the robustness and adaptability of the model in real-world scenarios. The experimental results demonstrate that the enhanced YOLOv7-tiny model achieves a mAP of 93.41%, which is a 2.98% improvement over the original model. The model also achieves precision and recall rates of 90.44% and 87.81%, respectively, with a FPS rate of 23.31. These results highlight the model’s excellence in both precision and real-time performance, making it highly effective for remote sensing aircraft detection and fine-grained feature recognition, particularly in complex environments where real-time processing is crucial.
To address the challenges of complex backgrounds, varying target scales, and dense targets in license plate detection within surveillance scenarios, we propose an enhanced method based on an improved YOLOv8n network. This approach involves redesigning key components of the YOLOv8n architecture, including the C2f module, the SPPF module, and the detection head. Additionally, we optimize the WIoU loss function, replacing the original CIoU loss function, which leads to improved bounding box feature extraction and enhanced regression accuracy. To evaluate the model’s robustness in complex environments with varying lighting, backgrounds, angles, and vehicle types, we created a custom surveillance license plate dataset. Experimental results show that the improved model achieves a notable increase in detection accuracy, with mAP@0.5 rising from 90.9% in the baseline model to 94.4%, precision improving from 90.2% to 92.8%, and recall increasing from 82.9% to 87.9%. Additionally, the model’s parameters are reduced from 3.1 M to 2.1 M, significantly enhancing computational efficiency. Moreover, the model achieves an inference speed FPS of 86, maintaining high precision and meeting real-time detection requirements. This demonstrates that our method provides an efficient and reliable solution for license plate detection in surveillance scenarios.
For vehicle driving scenarios in complex backgrounds, road sign detection faces challenges such as multi-scale targets, long-distances, and low-resolution. To address these challenges, a detection method based on an improved YOLO11n network is proposed. Firstly, to accommodate the multi-scale characteristics of the targets and improve the network’s ability to detect low-resolution objects and details, a Multi-path Gated Aggregation (MGA) Module is proposed, achieving these objectives via multi-dimensional feature extraction. Secondly, the Neck is improved by designing a network structure that incorporates high-resolution information from the Backbone, thereby enhancing the detection capabilities for small and blurry targets. Finally, an enhanced Spatial Pyramid Pooling—Fast (SPPF) module is proposed, wherein a Group Convolution-Layer Normalization-SiLU structure is integrated across various stages of information passing. By fusing adjacent channel information, it effectively suppresses complex background noise across multiple scales and amplifies road marking features, which consequently boosts the model’s discriminability for distant and obscured targets. Experimental results on a multi-type road sign dataset show that the improved model achieves an mAP@0.5 of 96.96%, which is 1.42% higher than the original model. The mAP@0.5–0.95 and Recall rates are 83.94% and 92.94%, respectively, while the inference speed remains at 134 FPS. Research demonstrates that via targeted modular designs, the proposed approach strikes a superior balance between detection accuracy and real-time efficiency. Consequently, it provides robust technical support for the reliable operation of intelligent vehicle perception systems under complex conditions.
In cotton disease detection, the complex farmland environment and the varying scales of disease spots, especially the presence of small-target disease spots, limit the detection accuracy of lightweight models. To address this issue, an improved YOLOv11n detection algorithm is proposed. First, the backbone network is reconstructed using the GhostConv (G-conv) module, which generates redundant feature maps through linear operations, thereby reducing computational complexity. Second, an Adaptive Calibration and Feature Fusion Architecture Head (ACFFA) with prior calibration and cross-scale fusion capabilities is constructed in the detection stage to handle the problem of varying disease spot scales. Furthermore, the Adaptive Scale-aware Wise Intersection over Union (AS-WIoU) loss function, improved from WIoUv3, is introduced to enhance the stability of bounding box regression and improve detection accuracy for low-resolution, small-target lesions. Experimental results show that on the cotton disease dataset constructed based on the Mendeley Data database, the proposed model achieves mAP50 and mAP50-95 of 90.30% and 73.84%, respectively, with precision and recall of 92.33% and 87.68%, and a parameter count of 3.81 M. The algorithm significantly improves detection accuracy while maintaining efficient inference, making it suitable for real-time monitoring tasks on agricultural embedded terminals.