Automated Intelligent Drip Monitoring System and Method

Abstract

This invention relates to an automated intelligent drip monitoring system, comprising: a camera for capturing images of the drip container and acquiring image data; a control module for controlling the camera's shooting conditions and receiving image data; and an edge computing device including an image recognition module for analyzing the image data using a deep learning model. The image recognition module consists of at least three layers of models, including: a feature labeling model for labeling the region of interest (ROI) corresponding to the drip container in the image data, generating a label file, and cropping the ROI feature image from the label file based on the ROI; a classification model for classifying the ROI feature image to obtain a classification result; and a percentage model for predicting the percentage of remaining liquid based on the classification result to obtain a predicted liquid level.

Claims

1. An automated intelligent drip monitoring system, comprising: a camera for capturing images of a drip container and acquiring image data; a control module connected to the camera for controlling the camera's shooting conditions and receiving the image data; an edge computing device connected to the control module, the edge computing device comprising an image recognition module for classifying the image data using a deep learning model, predicting and outputting the liquid level remaining in the drip container; the image recognition module comprising at least three layers, including: a feature labeling model for labeling the region of interest (ROI) in the image data corresponding to the drip container, generating a labeling file, and cropping the ROI feature image from the labeling file based on the region of interest; A classification model, connected to the feature labeling model, determines the capacity category of the drip container in the ROI feature image based on multiple image features of the drip container, outputs the corresponding category index number, and thus obtains the classification result; and multiple percentage models, each with its own weight, are connected to the classification model. The corresponding percentage model is selected based on the classification result to perform regression prediction of the remaining liquid volume and outputs the predicted liquid volume result. The multiple image features of the drip container include the bottle's external proportions, scale distribution features, bottle mouth and bottom structure, and overall size relative proportions.

2. The automated intelligent drip monitoring system as described in claim 1, wherein, The control module includes a microcomputer device selected from at least one of the following: Raspberry Pi, Orange Pi, Banana Pi, BeagleBone Black, and Tinker Board.

3. The automated intelligent drip monitoring system as described in claim 1, wherein, The edge computing device is a Jetson Nano.

4. The automated intelligent drip monitoring system as described in claim 1, wherein, The deep learning model used in the feature labeling model, the classification model, and the percentage model is the MobileNet model.

5. The automated intelligent drip monitoring system as described in claim 1, wherein, This feature labeling model uses multiple image augmentation methods to augment the ROI feature image.

6. The automated intelligent drip monitoring system as described in claim 1, wherein, The last layer of this classification model and the percentage model is a linear mapping layer.

7. The automated intelligent drip monitoring system as described in claim 1, wherein, The system further includes a weight sensing module connected to the drip container and the control module to measure the weight of the drip container and then calculate the percentage of liquid remaining.

8. The automated intelligent drip monitoring system as described in claim 1, wherein, The system further includes an alert module connected to the edge computing device. When the percentage shown by the predicted liquid level is lower than a set value, an alert message is generated to notify the user to replenish the IV fluid.

9. The automated intelligent drip monitoring system as described in claim 1, wherein, The control module further includes an exception handling module that automatically eliminates the abnormal state when the image data access fails.

10. An automated intelligent drip monitoring method, comprising: acquiring image data by photographing a drip container with a camera; classifying the image data using a deep learning model, predicting and outputting the remaining liquid level in the drip container; the image recognition process includes at least three deep learning-based procedures: a feature labeling procedure, which labels the region of interest (ROI) in the image data corresponding to the drip container, generates a label file, and crops the ROI feature image based on the ROI; a classification prediction procedure, which determines the capacity type of the drip container in the ROI feature image based on multiple image features of the drip container in the ROI feature image, outputs the corresponding category index number, and thus obtains the classification result; and a percentage prediction procedure, which uses multiple percentage models, each with its own weight, selects the corresponding percentage model based on the classification result, performs regression prediction of the remaining liquid level, and outputs the liquid level prediction result; wherein... The multiple image features of the drip container include the bottle's external proportions, scale distribution features, bottle mouth and bottom structure, and overall size relative proportions.

Images