Unmanned aerial vehicle distribution network inspection method with airborne front-end identification model
A technology for identifying models and unmanned aerial vehicles, which is applied in neural learning methods, character and pattern recognition, and biological neural network models. Robustness, ensuring the safety of distribution network transmission, and reducing computing costs
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0127] Example 1 Evaluation of the effect of the deep convolutional neural network transmission line defect recognition model
[0128] In order to ensure that the dual-view images input into the network correspond to the same target, two drones take off and fly synchronously along the two sides of the equipment to be inspected at the same time. The exposure is controlled by the same system. The shooting angles of the two cameras are perpendicular to each other. images to ensure that the bi-view images fed into the network correspond to the same object.
[0129] Each convolutional neural network has a total of 3 convolutional layers: ①The first layer contains 28 convolutional kernels, the size of which is 21×21, and the sliding window is performed with an interval of 1, and the ReLU activation function is used;② The second layer contains 32 convolution kernels, the size of the convolution kernel is 11×11, the sliding window is performed at an interval of 1, and the ReLU activat...
Embodiment 2
[0136] Example 2 Evaluation of the effect of the dual-view convolutional neural network defect classification model
[0137] In the experiment, a total of 2048 defect cases were used to train the dual-view convolutional neural network defect classification model, including 623 cases of line break defects, 518 cases of switch tripping, 773 cases of equipment corrosion, and 134 cases of other defects. In addition, 1254 defect samples were used to test the effect of the defect classification model, including 262 cases of fracture defects, 481 cases of switch tripping, 274 cases of equipment corrosion, and 237 cases of other defects. Transmission line defect discrimination is a multi-classification problem. The present invention adopts the strategy of One-vs-rest to build a multi-classification model. The method is to classify certain types of defects into one class of samples and the rest of the defects into another class of samples when training the model. , so that four types o...
Embodiment 3
[0147] Embodiment 3 The effectiveness of fixed-point adaptive selection convolutional neural network
[0148] In the transmission line defect image classification experiment, there are 45,000 training samples and 5,000 test samples. The model was tested from the aspects of running time, energy consumption, and classification accuracy, which proved the effectiveness of the fixed-point adaptive selection convolutional neural network. The test results are as follows: Figure 20 shown.
[0149] In terms of inference time, by comparing the single models MobileNet_28, Inceptionv_4_58 and ResNet_v1_152 with the multi-level model KNN_based, it can be found that MobileNet_28 has the fastest computing speed and the lowest accuracy rate. Although the running time of the multi-level model KNN_based is higher than ResNet_v1_152, compared to Inception_4_58 has decreased, and KNN_based has the highest accuracy. Therefore, the adaptive selection strategy of the multi-level image classificat...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More