摘要 影像辨識是工業視覺檢測的核心技術，而照明和色溫經常是影響辨識性能最主要的原因之一，正確的色溫估測可以有效提升影像辨識性能。傳統的解決方案主要關注於在特定光源下如何透過演算法補償圖像色度。本研究設計了一個智慧化校正色溫的視覺感測器，藉由光譜感測器的六個光譜反射訊號，建立一個神經網路色彩校正模型，在不同的照明環境下進行取像，可提供良好的色彩恆定性。我們在ARM Cortex-M7的嵌入式平台實現了取像、光譜量測、神經網路色彩校正、和打光控制功能，完成了一個可自動校正色溫的智慧化視覺感測系統。最後，我們以咖啡豆烘培程度檢測，來驗證此一視覺感測系統。本系統不需透過複雜的演算法補償圖像色度亦不需拍攝多張圖像比較差異，僅需在使用前實施環境色溫的校正即可達到優異的影像辨識結果。對於色差變異嚴重的檢測環境，我們的系統也可以達到良好的檢測的可靠度，可以應用在廣泛的視覺檢測。;Abstract Image recognition is the core technology of industrial vision detection, and lighting and color temperature are often one of the most important reasons to affect identification performance. Correct color temperature estimation can effectively improve image recognition performance. The traditional solution focuses on how to compensate the image chroma through the algorithm under a specific light source. In this study, an intelligent visual sensor with positive color temperature is designed, and a neural network color correction model is established by six spectral reflection signals of spectral sensors, which can provide good color constancy by image in different illumination environments. We have realized the functions of image, spectral measurement, neural network color correction, and light control in the embedded platform of ARM Cortex-M7, and completed an intelligent visual sensing system which can automatically and more positive color temperature. Finally, we verify this visual sensing system by testing the baking degree of coffee beans. This system does not need to compensate the image chroma through the complex algorithm also does not need to take multiple images to compare the difference, only needs to implement the ambient color temperature correction before the use can achieve excellent image identification results. For the detection environment with serious chromatic aberration variation, our system can also achieve good reliability of detection, which can be applied to a wide range of visual inspection.
