研究目的
To automate the generation of solar panel layouts by detecting obstacles to their exact edges using a fusion of object detection and edge detection algorithms, reducing manual effort and improving accuracy.
研究成果
The proposed fusion of object detection and edge detection algorithms effectively automates solar panel layout generation, achieving high accuracy with edge pixel count variation under 25%. This approach is applicable to various domains beyond solar panels, such as medical imaging and drone navigation. Future work should focus on improving speed and exploring other detector combinations.
研究不足
The evaluation metrics (edge pixel count and variance) are subjective and not the best measures; false positives and negatives may occur, but user correction is allowed. The framework may not handle all types of obstacles perfectly, and speed could be improved for real-time applications.
1：Experimental Design and Method Selection:
The framework combines object detection APIs (e.g., TensorFlow Object Detection API) with traditional edge detection algorithms (Prewitt, Sobel, Laplacian, Canny) to achieve fine object detection. The rationale is to first detect coarse objects and then apply edge detection on candidate regions for precise boundary identification.
2：Sample Selection and Data Sources:
Roof images were obtained from PVComplete dataset, which includes JSON objects with latitude, longitude, and obstacle coordinates. Images were downloaded using Google Maps API and manually annotated with 'labelImg' tool.
3：List of Experimental Equipment and Materials:
A computer system with Python for scripting, TensorFlow framework, pre-trained MobileNet v1 model, and edge detection algorithms implemented in software.
4：Experimental Procedures and Operational Workflow:
Steps include training a CNN model on annotated roof images using transfer learning with MobileNet v1, performing coarse object detection to generate bounding boxes, applying edge detection on each candidate region, and fusing results back to the original image for visualization.
5：Data Analysis Methods:
Evaluation based on edge pixel count comparison with ground truth (manual annotations) and variance in edge lengths. Statistical analysis used to measure performance, with Canny edge detector showing less than 25% variation on average.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
