Application of Transfer Learning-Based ResNET50_CBAM In The Classification and Recognition of Drilling Fluid Flocculation Images

Drilling fluid is an important material in oil drilling, and there is a lack of effective identification method for the flocculation identification effect in the reuse of drilling fluid, and the flocculation effect of drilling fluid is judged by field workers according to experience, and this method is inefficient and inaccurate. In this paper, the initial ResNet50 network model is used to identify the flocculation images of drilling fluid by adding the attention mechanism CBAM to the initial ResNet50 network model, and the improved network model is called ResNet50_CBAM. In this paper, the flocculation images are enriched by data augmentation, and then the ResNet50_CBAM network model parameters are finetuned to extract features based on transfer learning, and the flocculation images are classified by softmax. At the same time, the evaluation indexes of other classical network models AlexNet, VGG16, GoogleNet, ResNet50 and MobileNetV3 in terms of overall accuracy, precision, recall and specificity were compared, and the ResNet50_CBAM network model was better than other network models, with an overall accuracy of $98.9 \%$.