Design Of Copper Composition Soft-sensing System Based On DaVinci Platform

For specific occasions where soft-sensing techniques are needed, it’s a big challenge to choose reasonable hardware architecture, and to design efficient software framework or algorithm with high performance. Regularly, the measurement of copper composition is behind-time, to solve this, a soft measurement system is designed and realized based on TMS320DM6446(DaVinci family, ARM and DSP double-core processor) in this article. Regarding real-time, accuracy as the goal, the article designs an efficient software framework, and puts forward a series of high-performance algorithms:SBRE method to extract region of interest(ROI) of the copper image under complex background, ROI quality detection method based on texture, ROI color feature extraction method, and S-GMR soft-sensing regression method. All these algorithms and the framework have been successfully realized on the DaVinci platform. The test demonstrates that the system is fast-speed and accurate.This article describes the whole work in detail, the main contributions and innovations are as follows:1. Construction of DaVinci dual-core hardware and software platform, and design of a high-performance software framework:ARM equipped with MontaVista Linux(a real-time operating system), uses multi-threaded framework to achieve image acquisition, processing, display, human-computer interaction and other functions; DSP equipped with DSP/BIOS(a real-time multi-tasking operating system), is responsible for image processing algorithms and soft-sensing operations. Thus, two cores will give full play to their respective advantages.2. ROI extraction:The article puts forward SBRE method to extract region of interest(ROI) of the copper image under complex background, which has good robustness and accuracy.3. ROI quality detection and feature extraction:Since the ROI of copper image contains texture noise, a method which combines LBP, VAR and gray variance is put forward to detect the ROI quality. Obviously, extracting the feature(average of color-vector-angle) after ROI quality detection can ensure the accuracy of copper composition estimation later.4. Research and realization of the copper composition regression model:For the estimation problem of the copper composition, S-GMR method based on support vector machine classification and Gaussian mixture regression is put forward. It uses Gaussian function to simulate the probability distribution of the copper composition and the average of color-vector-angle, which may contain reasonable physical meaning. Compared with the ε-SVR method, S-GMR shows its advantages. Finally S-GMR is applied to the DaVinci platform and the test demonstrates excellent real-time performance and accuracy.