目的 为了提高生产线上生物芯片点样质量检测的精度与效率，研究基于图像处理和卷积神经网络的算法，判断某生物芯片点样质量是否合格，并检测点样合格的生物芯片上的点样点半径。方法 采用CCD相机获取生物芯片点样后的图像，通过图像预处理，利用canny边缘检测和圆的拟合等图像处理方法，得到点样点的几何信息，进而计算出点样点半径。同时提出基于卷积神经网络的点样质量检测方法，通过区域建议网络提取点样点卷积特征，引入分割全连接层来训练检测模型，通过离线训练来验证获得模型的最佳参数。结果 和手动测量结果进行对比发现，半径误差不超过±0.1 mm，点样质量检测准确率为91.1%，单个生物芯片检测时间总和不超过1.6 s。结论 所提出的方法能够满足生产线上产品检测准确性和实时性的要求。

In order to improve the precision and efficiency of biochip sample application quality inspection on the production line, the work uses a method based on image processing and convolutional neural network to determine whether a biochip sample application quality is acceptable and inspect the radius of sample application point on the qualified biochip. The CCD camera was used to obtain the image after the biochip was sampled. Through image preprocessing, the image processing methods such as canny edge detection and round fitting were used to obtain the geometric information of the spotted point, and then the sample application point’s radius was calculated. At the same time, the detection method of sample application’s quality based on convolutional neural network was proposed. The convolution features of sample application points were extracted by the regional recommendation network, and the detection model was introduced by dividing the fully connected layer. The offline training was used to verify the method to obtain the best parameters of the model. Compared with the manual measurement results, the radius error did not exceed ±0.1mm, and the accuracy rate of sample application quality detection was 91.1%. The total detection time of a single biochip did not exceed 1.6 seconds. The proposed method can meet the requirements of accuracy and real time of product testing on the production line.