对自动显微镜的自动聚焦评价函数及聚焦控制策略进行了研究。首先, 介绍了频域聚焦函数提升小波变换及时域聚焦函数Sobel-Tenengrad算子, 通过将提升小波变换和Sobel-Tenengrad算子有机组合提出了一种新型聚焦评价函数。然后, 利用离焦、正焦样本图像对自组织算法进行无监督训练, 使用粒子群优化算法加速训练过程, 并以经过学习的自组织映射算法作为聚焦控制器。最后, 进行了显微视觉自动聚焦实验。实验结果表明：新型组合算子具有单峰性, 峰值处变化陡峭, 对不同样本、不同倍数物镜均可在正焦位置达到最大值, 鲁棒性强; 经过学习控制器后平均仅用7.6步即可完成自动聚焦, 与爬山法相比, 该聚焦算法不仅大大提高了聚焦速度且性能稳定, 对每幅输入图像处理、识别时间约为120 ms; 满足了显微视觉自动聚焦要求, 获得了良好聚焦效果。

The auto-focusing evaluation function and a controller were researched based on an automated microscopy. At first, the Discrete Wavelet Transform (DWT) and Sobel-Tenengrad function were introduced, and a new auto-focusing evaluation function was proposed by combining the DWT with the Sobel-Tenengrad operator. Then, the defocused and focused sample pictures were used to train the Self-organizing Map (SOM) algorithm in a unsupervised method, and the Particle Swarm Optimization (PSO) was used to accelerate the training process. Finally, an auto-focusing experiment was carried out by using the trained SOM controller. The experimental results show that the new auto-focusing function has the characters of single steep peak and strong robustness to different samples and objective lenses. The results also indicate that the SOM based controller only takes 7.6 steps for auto-focusing process on average, and the focusing speed and stability have been greatly improved compared with that using the mountain climbing method. Moreover, it processes or recognizes the input image only for about 120 ms. The proposed method has met the requirements of auto-focusing of micro-vision system, and obtained good results.