In the process of scene based non-uniformity correction, the scene is often changeable, and the corresponding detector operating range of different temperature scenes is different, and their nonuniformity parameters are also different, so when the scene changes, the non-uniformity parameters will also drift, and it is easy to introduce serious ghosts in the correction results, which greatly reduces the Rate of convergence of the algorithm, This poses serious problems for real-time non-uniformity correction algorithms. At the same time, the scene is very flexible, and a single model cannot describe the scene well. Therefore, on the basis of treating the dynamic change process of the scene as a Markov process, this article proposes an adaptive non-uniformity correction algorithm based on Multi Model Particle Filter (PF-NUC). By introducing a tracking framework of particle filter, a nonlinear and non Gaussian parameter estimation model is established. Through experimental simulation verification, after the algorithm proposed in this article, almost no ghosts or residual non-uniformity can be seen. Through visual evaluation, the PF-NUC method has the best ability to remove fixed pattern noise.
A non-uniformity correction method based on machine learning for A/D driver circuit level. Firstly, different types of infrared detectors are placed in a temperature uniform radiation field. When they are working normally, the analog output signal waveform of each detector in multiple scenarios is collected multiple times to obtain the approximate average voltage value of each line in a frame, and saved as a document; Secondly, using GPU for machine learning of the above documents and accurately driving the D/A chip for digital to analog conversion, simulating the waveform voltage value of the analog output signal mentioned above to generate waveform voltage with similar nonuniformity; Thirdly, the voltage waveform generated by the multi-channel voltage output digital to analog converter is followed by an operational amplifier, filtered, and then output to the A/D chip as the reference voltage for sampling; Finally, the analog video signal output by the infrared detector is sampled and quantized by an A/D chip to obtain a more uniform image digital signal.
The target detection ability of external thermal imaging system is affected by many factors, including the source and influence of system heterogeneity, the digital enhancement mechanism of infrared imaging, the realizability and real-time of system super-resolution and so on. The infrared focal plane detector integrates focal plane daylighting array, signal readout circuit, temperature control circuit, timing control circuit, etc. requires the external supply of appropriate driving voltage and digital timing signal to work normally. Because the infrared detector belongs to the front-end device of the system with high sensitivity, its requirements for working voltage and digital timing are also relatively strict. The quality of driving signal determines whether the infrared detector can work in the best state, so that the following image processing circuit can get the best original infrared image. By studying the imaging principle of uncooled focal plane detector, this paper puts forward the key factors affecting the performance of infrared image, and puts forward hardware based low-noise bias voltage signal generation technology, which improves the intelligent level of the system under the condition of ensuring the stability of the system.
In order to suppress image ripples of the uncooled infrared detector, which is caused by the bias voltage noise bought by the un-uniform characteristic of column channel in the focal plane and changed varying with time, a hardware solution by structure extending is given including the principle and simulation of the design. As a material which is not sensitive to the environmental radiation temperature, the sapphire is used to extend the focal plane as the “background array”, while the readout circuit (ROIC) also extending same pixels. Then the noise value caused by the bias voltage can be calculated by averaging the value of the background pixels, which is subtracted from the sampling value of original array. Simulation results indicate that RMSE of one video image falls from 115 to 10 after 20 frames, while PSNR increases from 30 to 105. It can get expected IR images sustained at a uniformity level and the bias voltage ripples of images are suppressed effectively.
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.