Aiming at the problem that optical-image-based methods are susceptible to lighting conditions and have difficulty identifying oil films

a method for identifying floating objects on water surface based on millimeter-wave radar combined with deep learning models is proposed.Data collection of indoor water surface scenes and floating objects on outdoor lake surface is completed.By introducing multiple deep learning models

it is found by comparison that the second-order convolutional neural network(CNN)model combined with time-domain graphs achieves the best performance

with a recognition accuracy of 97.68%.Aiming at the problem of insufficient convolutional layers in the second-order-CNN model

an attention mechanism and a batch normalization layer are introduced into the model

and a fusion algorithm model named Attention-CNN_bn is proposed.By comparing the Attention-CNN_bn model with the CNN model

the results show that the combination of the timedomain graph and the AttentionCNN_bn model is the best recognition scheme

with a recognition accuracy of 98.39%.The accuracy of this scheme for oil film recognition is 98.75%.Compared with image recognition

which has difficulty in identifying oil films

millimeter-wave radar achieves accurate recognition of oil films and is not affected by lighting conditions.