Rock image classification is a significant part of geological research. Compared with traditional image classification methods, rock image classification methods based on deep learning models have the great advantage in terms of automatic image features extraction. However, the rock classification accuracies of existing deep learning models are unsatisfied due to the weak feature extraction ability of the network model. In this study, a deep residual neural network (ResNet) model with the transfer learning method is proposed to establish the corresponding rock automatic classification model for seven kinds of rock images. ResNet34 introduces the residual structure to make it have an excellent effect in the field of image classification, which extracts high-quality rock image features and avoids information loss. The transfer learning method abstracts the deep features from the shallow features, and better express the rock texture features for classification in the case of fewer rock images. To improve the generalization of the model, a total of 3,82,536 rock images were generated for training via image slicing and data augmentation. The network parameters trained on the Texture Library dataset which contains 47 types of texture images and reflect the characteristics of rocks are used for transfer learning. This pre-trained weight is loaded when training the ResNet34 model with the rock dataset. Then the model parameters are fine-tuned to transfer the model to the rock classification problem. The experimental results show that the accuracy of the model without transfer learning reached 88.1%, while the model using transfer learning achieved an accuracy of 99.1%. Aiming at geological engineering field investigation, this paper studies the embedded deployment application of the rock classification network. The proposed rock classification network model is transplanted to an embedded platform. By designing a rock classification system, the off-line rock classification is realized, which provides a new solution for the rock classification problem in the geological survey. The deep residual neural network and transfer learning method used in this paper can automatically classify rock features without manually extracting. These methods reduce the influence of subjective factors and make the rock classification process more automatic and intelligent.

Rock image classification is a fundamental and crucial task in the creation of geological surveys. Traditional rock image classification methods mainly rely on manual operation, resulting in high costs and unstable accuracy. While existing methods based on deep learning models have overcome the limitations of traditional methods and achieved intelligent image classification, they still suffer from low accuracy due to suboptimal network structures. In this study, a rock image classification model based on EfficientNet and a triplet attention mechanism is proposed to achieve accurate end-to-end classification. The model was built on EfficientNet, which boasts an efficient network structure thanks to NAS technology and a compound model scaling method, thus achieving high accuracy for rock image classification. Additionally, the triplet attention mechanism was introduced to address the shortcoming of EfficientNet in feature expression and enable the model to fully capture the channel and spatial attention information of rock images, further improving accuracy. During network training, transfer learning was employed by loading pre-trained model parameters into the classification model, which accelerated convergence and reduced training time. The results show that the classification model with transfer learning achieved 92.6% accuracy in the training set and 93.2% Top-1 accuracy in the test set, outperforming other mainstream models and demonstrating strong robustness and generalization ability.

It is challenging to accurately detect flexible objects with arbitrary orientation from monitoring images in power grid maintenance and inspection sites. This is because these images exhibit a significant imbalance between the foreground and background, which can lead to low detection accuracy when using a horizontal bounding box (HBB) as the detector in general object detection algorithms. Existing multi-oriented detection algorithms that use irregular polygons as the detector can improve accuracy to some extent, but their accuracy is limited due to boundary problems during the training process. This paper proposes a rotation-adaptive YOLOv5 (R_YOLOv5) with a rotated bounding box (RBB) to detect flexible objects with arbitrary orientation, effectively addressing the above issues and achieving high accuracy. Firstly, a long-side representation method is used to add the degree of freedom (DOF) for bounding boxes, enabling accurate detection of flexible objects with large spans, deformable shapes, and small foreground-to-background ratios. Furthermore, the further boundary problem induced by the proposed bounding box strategy is overcome by using classification discretization and symmetric function mapping methods. Finally, the loss function is optimized to ensure training convergence for the new bounding box. To meet various practical requirements, we propose four models with different scales based on YOLOv5, namely R_YOLOv5s, R_YOLOv5m, R_YOLOv5l, and R_YOLOv5x. Experimental results demonstrate that these four models achieve mean average precision (mAP) values of 0.712, 0.731, 0.736, and 0.745 on the DOTA-v1.5 dataset and 0.579, 0.629, 0.689, and 0.713 on our self-built FO dataset, exhibiting higher recognition accuracy and a stronger generalization ability. Among them, R_YOLOv5x achieves a mAP that is about 6.84% higher than ReDet on the DOTAv-1.5 dataset and at least 2% higher than the original YOLOv5 model on the FO dataset.

In the operation site of power grid construction, it is crucial to comprehensively and efficiently detect violations of regulations for the personal safety of the workers with a safety monitoring system based on object detection technology. However, common general-purpose object detection algorithms are difficult to deploy on low-computational-power embedded platforms situated at the edge due to their high model complexity. These algorithms suffer from drawbacks such as low operational efficiency, slow detection speed, and high energy consumption. To address this issue, a lightweight violation detection algorithm based on the SP (Segmentation-and-Product) attention mechanism, named SP-YOLO-Lite, is proposed to improve the YOLOv5s detection algorithm and achieve low-cost deployment and efficient operation of object detection algorithms on low-computational-power monitoring platforms. First, to address the issue of excessive complexity in backbone networks built with conventional convolutional modules, a Lightweight Convolutional Block was employed to construct the backbone network, significantly reducing computational and parameter costs while maintaining high detection model accuracy. Second, in response to the problem of existing attention mechanisms overlooking spatial local information, we introduced an image segmentation operation and proposed a novel attention mechanism called Segmentation-and-Product (SP) attention. It enables the model to effectively capture local informative features of the image, thereby enhancing model accuracy. Furthermore, a Neck network that is both lightweight and feature-rich is proposed by introducing Depthwise Separable Convolution and Segmentation-and-Product attention module to Path Aggregation Network, thus addressing the issue of high computation and parameter volume in the Neck network of YOLOv5s. Experimental results show that compared with the baseline network YOLOv5s, the proposed SP-YOLO-Lite model reduces the computation and parameter volume by approximately 70%, achieving similar detection accuracy on both the VOC dataset and our self-built SMPC dataset.

Slender flexible objects are ubiquitous in real-world circumstances. The existing object detection and segmentation algorithms have achieved high accuracy and speed in rigid objects, but the detection effect of slender flexible objects is not ideal. Extreme aspect ratios and dynamically changeable geometric appearance characterize slender flexible objects, to the extent that it is difficult to locate them in instance segmentation. In this paper, a new instance segmentation method based on the object correlation module and loss function optimization is proposed for the detection of slender flexible objects. In order to achieve more accurate anchor box positioning, a GIoU bounding-box regression loss function is selected to overcome the problem of inconsistency between training objectives and assessment indicators. Furthermore, due to the Mask Scoring R-CNN detection network ignoring the relationships between objects, the object correlation module is proposed to achieve end-to-end learning and modeling of the correlation features between all objects in the image to improve slender flexible objects detection accuracy. The results of the experiments on the self-built flexible object dataset for the power grid operation sites demonstrate that the method presented in this research can efficiently recognize and segment slender flexible objects, with a detection accuracy of 44.8%. The ablation experiment also shows that the addition object correlation module and the revised bounding-box regression loss function are both effective and can enhance slender flexible object detection accuracy by 1.2% and 0.5%, respectively. The proposed instance segmentation method considers the correlation characteristics between objects and improves the bounding-box regression loss function to increase the segmentation accuracy of slender flexible objects.

In power inspection, it is crucial to accurately and regularly monitor the status of isolation switches to ensure the stable operation of power systems. However, current methods for detecting the open and closed states of isolation switches based on image recognition still suffer from low accuracy and high edge deployment costs. In this paper, we propose a lightweight object detection model, EMB-YOLO, to address this challenge. Firstly, we propose an efficient mobile inverted bottleneck convolution (EMBC) module for the backbone network. This module is designed with a lightweight structure, aimed at reducing the computational complexity and parameter count, thereby optimizing the model’s computational efficiency. Furthermore, an ELA attention mechanism is used in the EMBC module to enhance the extraction of horizontal and vertical isolation switch features in complex environments. Finally, we proposed an efficient-RepGDFPN fusion network. This network integrates feature maps from different levels to detect isolation switches at multiple scales in monitoring scenarios. An isolation switch dataset was self-built to evaluate the performance of the proposed EMB-YOLO. The experimental results demonstrated that the proposed method achieved superior detection performance on our self-built dataset, with a mean average precision (mAP) of 87.2%, while maintaining a computational cost of only $6.5\times {10}^9$ FLOPs and a parameter size of just $2.8\times {10}^6$ bytes.