Remote sensing image target detection method based on deep neural network
A deep neural network and remote sensing image technology, which is applied in the field of digital image processing and pattern recognition, can solve the problems of insufficient feature extraction capability of shallow CNN model, inaccurate detection results of remote sensing image targets, and inability to fine-tune deep CNN models. Smaller targets and complex backgrounds, reduced manual labeling costs, and the effect of omitting screening
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Example Embodiment
[0056] Example 1
[0057] See Figure 1a-1b The test sample image in this embodiment comes from the Satellite2000 remote sensing image data set. The test sample image in the Satellite2000 remote sensing image data set is generally a small part of the airport, including 2-8 aircraft. The present invention executes the aircraft on the Satellite2000 remote sensing image data set. The task of detection. The size range of the test sample image is 256×256 to 500×500, and the test sample image or part of it does not appear in the training sample.
[0058] See Figure 4 The remote sensing image target detection method based on the deep neural network of this embodiment consists of two steps: training the detection model and testing the detection model. The steps of training the detection model are as follows:
[0059] (1) Obtain training samples and preprocess them
[0060] (a) Select 1,000,000 common object sample images from the daily common object data set ILSVRC-2012 (Large Scale Visual ...
Example Embodiment
[0093] Example 2
[0094] See figure 2 The test sample image in this embodiment comes from the Satellite Aircrafts Dataset remote sensing image data set. The test sample image in the Satellite Aircrafts Dataset remote sensing image data set is generally a larger part of the airport, including 10-20 aircraft. Perform aircraft detection tasks on the image data set. The size range of the test sample image is 300×300 to 800×800, and the test sample image or part of it does not appear in the training sample.
[0095] See Figure 4 The remote sensing image target detection method based on the deep neural network of this embodiment consists of two steps: training the detection model and testing the detection model. The steps of training the detection model are as follows:
[0096] (1) Obtain training samples and preprocess them
[0097] Obtaining training samples and performing preprocessing are the same as in Embodiment 1;
[0098] (2) Label training samples
[0099] The marked training sa...
Example Embodiment
[0118] Example 3
[0119] See image 3 The test sample image in this embodiment comes from the Aircrafts Dataset remote sensing image data set. The test sample image in the Aircrafts Dataset remote sensing image data set generally covers the entire airport area, including 30-50 aircraft. The present invention is implemented on the Aircrafts Dataset remote sensing image data set. The task of aircraft inspection. The size range of the test sample image is 800×800 to 1400×1400, and the test sample image or part of it does not appear in the training sample.
[0120] See Figure 4 The remote sensing image target detection method based on the deep neural network of this embodiment consists of two steps: training the detection model and testing the detection model. The steps of training the detection model are as follows:
[0121] (3) Obtain training samples and preprocess them
[0122] Obtaining training samples and performing preprocessing are the same as in Embodiment 1;
[0123] (4) Lab...
PUM
Abstract
Description
Claims
Application Information
- R&D Engineer
- R&D Manager
- IP Professional
- Industry Leading Data Capabilities
- Powerful AI technology
- Patent DNA Extraction
Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.
© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap