Object-Based Flow Map Classification Combining Speckle Imaging and Machine Learning

The study of the relationship between the shape of an object and the patterns of liquid flowing through them is of great interest in many fields of science and industry. Generally, objects with different geometry will generate different flow map which is related to their shape. A wide variety of methods have been proposed with the goal of detecting object by shapes. In this context, here, we explore and demonstrate the combination of machine learning and laser speckle contrast imaging to classify the shape of an object via liquid flowing through it. In our setup, laser light shines on the flow which passes through four objects of different shapes. The diffused speckle images are acquired by a camera and are converted to flow maps on the computer. These maps are then fed into a convolutional neural network for classification and recognition of the objects. We created a database with experimental flow maps and trained the SqueezeNet model to classify new maps; 70% images were used for training and 30% for validation and testing per object. Through experimental results, we show that the proposed method can successfully classify objects with a high accuracy rate. The findings of this study could, for example, be used to discriminate between different types of prostate cancer, especially where identifying abnormal flow map during urination holds diagnostic value.