RESUMO
Rust of transmission line fittings is a major hidden risk to transmission safety. Since the fittings located at high altitude are inconvenient to detect and maintain, machine vision techniques have been introduced to realize the intelligent rust detection with the help of unmanned aerial vehicles (UAV). Due to the small size of fittings and disturbance of complex environmental background, however, there are often cases of missing detection and false detection. To improve the detection reliability and robustness, this paper proposes a new robust Faster R-CNN model with feature enhancement mechanism for the rust detection of transmission line fitting. Different from current methods that improve feature representation in front end, this paper adopts an idea of back-end feature enhancement. First, the residual network ResNet-101 is introduced as the backbone network to extract rich discriminative information from the UAV images. Second, a new feature enhancement mechanism is added after the region of interest (ROI) pooling layer. Through calculating the similarity between each region proposal and the others, the feature weights of the region proposals containing target object can be enhanced via the overlaying of the object's representation. The weight of the disturbance terms can then be relatively reduced. Empirical evaluation is conducted on some real-world UAV monitoring images. The comparative results demonstrate the effectiveness of the proposed model in terms of detection precision and recall rate, with the average precision of rust detection 97.07%, indicating that the proposed method can provide an reliable and robust solution for the rust detection.
Assuntos
Reprodutibilidade dos TestesRESUMO
To improve the motion distortion caused by LiDAR data at low and medium frame rates when moving, this paper proposes an improved algorithm for scanning matching of estimated velocity that combines an IMU and odometer. First, the information of the IMU and the odometer is fused, and the pose of the LiDAR is obtained using the linear interpolation method. The ICP method is used to scan and match the LiDAR data. The data fused by the IMU and the odometer provide the optimal initial value for the ICP. The estimated speed of the LiDAR is introduced as the termination condition of the ICP method iteration to realize the compensation of the LiDAR data. The experimental comparative analysis shows that the algorithm is better than the ICP algorithm and the VICP algorithm in matching accuracy.
RESUMO
The diagnosis of depression almost exclusively depends on doctor-patient communication and scale analysis, which have the obvious disadvantages such as patient denial, poor sensitivity, subjective biases and inaccuracy. An objective, automated method that predicts clinical outcomes in depression is essential for increasing the accuracy of depression recognition and treatments. This paper aims at better recognizing depression using the transformation of EEG features and machine learning methods. An experiment based on emotional face stimuli task was conducted, and twenty-eight subjects' EEG data were recorded from 128-channel HydroCel Geodesic Sensor Net (HCGSN) by Net Station software. The Mini International Neuropsychiatric Interview (MINI) was used by psychiatrists as the criterion for diagnosis of depression patients. The power spectral density and activity were respectively extracted as original features using Auto-regress model and Hjorth algorithm with different time windows. Two separate approaches processed the features: ensemble learning and deep learning. For the ensemble learning, a deep forest transformed the original features to new features that potentially improve feature engineering and a support vector machine (SVM) that was applied as classifier. For deep learning method, we added spatial information of EEG caps to both features by image conversion and adopted convolutional neural network (CNN) to recognize them. The performance of both methods was evaluated for separated and total frequency bands. As a result, the best accuracy obtained was 89.02% when we used the ensemble model and power spectral density. The best accuracy of deep learning method was 84.75% using the activity. These experimental results prove the efficiency of the proposed methods and show that EEG could be used as a reliable indicator for depression recognition, which makes it possible for EEG-based portable system design and application in auxiliary depression recognition in the future.