In view of the current image segmentation field, there are few studies on the segmentation of typical wire clamp components of transmission lines. Traditional image processing methods have low segmentation accuracy and require artificial design of feature extraction methods, which are usually only suitable for equipment of a certain structure with insufficient generalization. In this paper, an infrared image segmentation method based on Mask R-CNN (Mask region-based convolutional neural network) for typical guide-ground lines is proposed. Its structure takes Mask R-CNN model combined with FPN (Feature pyramid structure) as the basic framework, and uses RPN (Regional proposal network) to generate candidate regions. Features are extracted from each candidate region through RoI Align layer, and then connected to FC (Fully connected layer) to achieve target classification and bbox (bounding box) regression. A mask branch is also added to predict the segmentation mask. The design can integrate multi-scale and multi-level semantic information to improve the recognition rate when extracting image features. In addition, the network structure is optimized by single channel for infrared images to reduce the size of the model and make it more lightweight. Ablation experiments were performed on two GTX 2080Ti graphics cards to verify the effectiveness of the proposed structure, and the mAP (mean average accuracy) of 0.421 was achieved with an IoU (Intersection over Union) threshold of 0.5.
Bronze inscription is one of the earliest well-established writing systems dating back to Shang dynasty in China. The Recognition of Bronze character recognition plays an important role in the identification and interpretation of Bronze inscription which traditionally is a tough and challenging task. To deal with class imbalance of training data in bronze inscription recognition, we propose a method based on few-shot learning. The recognition process consists of three stages. In the first stage, a model is pretrained in a large-scale character dataset with a novel negative margin loss. In the second stage, the pretrained weights of the backbone network is transferred to the target dataset. In the final stage, the distribution of few-shot classes is calibrated and a new classifier is re-trained accordingly. Through qualitative and quantitative experimental analyses, the proposed method exceeds the state-of-the-art on our Bronze Character dataset.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.