Multimode fibers(MMF) play a crucial role in promoting the miniaturization of endoscopes. With the development of deep learning and machine learning, neural networks can be used to recognize and classify the speckle patterns obtained at the output of optical fibers. Based on the speckle pattern of HERLEV cell images transmitted by multimode fiber, this paper studies the recognition accuracy of support vector machines (SVM), k-nearest neighbors (KNN) and convolutional neural networks (Inception V3) for multiple types of speckle. The experimental results show that Inception V3 has the highest classification accuracy, which is 1.60% and 0.58% higher than the classification accuracy of SVM and KNN algorithms on 7 types of cervical cell data sets respectively, reaches 97.9% accuracy, confirming the effectiveness of Inception V3 algorithm in identifying speckle.