Semantic Segmentation Neural Network for Spill Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting substance spills on solid surfaces are inefficient and pose health and safety risks due to reliance on human observation, which is slow and exposes individuals to hazards, and machine learning models struggle to reliably identify spills due to their varied shapes and forms.
Innovation Solution
A computer-implemented method using a trained semantic segmentation neural network to classify pixels in images of solid surfaces, determining spill classification values to detect substance spills automatically, with the network trained on datasets annotated with spill presence, allowing for reliable detection regardless of spill shape.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a person manually detects substance spills by visual observation, then the detection can be performed with simple equipment, but the detection is slow and exposes the person to health and safety risks
Solution Approach 1:
The patent replaces manual visual detection with an automated image processing system using machine learning models. The system captures images of solid surfaces and uses trained neural networks to automatically identify substance spills, eliminating the need for human observation and significantly reducing detection time while maintaining high reliability.
2Extent of automation
If traditional object detection machine learning models are used to detect substance spills, then automation is achieved, but the models struggle to reliably identify spills due to their varied shapes and forms
Solution Approach 1:
The patent employs semantic segmentation neural networks that divide the image into multiple segments, classifying each pixel to determine whether it belongs to a substance spill or the solid surface. This pixel-level segmentation approach enables the model to accurately detect spills of any shape or form by analyzing individual pixel characteristics rather than relying on predefined object shapes.
Solution Approach 2:
The patent transforms the detection approach by changing from object-level detection to pixel-level classification. The neural network outputs a classification value for each pixel indicating the probability of it being a spill, allowing the system to reliably detect spills regardless of their shape, size, or configuration.
3Productivity
If automated image processing is implemented to detect substance spills, then detection speed and safety are improved, but the system complexity increases due to the need for trained neural networks and image processing infrastructure
Solution Approach 1:
The patent implements a training phase before deployment where the semantic segmentation neural network is trained on annotated images of solid surfaces with substance spills. This preliminary training action enables the model to learn spill patterns and characteristics, allowing it to rapidly and accurately detect spills during operational use without requiring complex real-time analysis.
Data Source
AI summary
System, apparatus and method of image processing to detect a substance spill on a solid surface such as a floor is disclosed. First data representing a first image, captured by an image sensor, of a region including a solid surface, is received. A trained semantic segmentation neural network is applied to the first image data to determine, for each pixel of the first image, a spill classification value associated with the pixel, the determined spill classification value for a given pixel indicating the extent to which the trained semantic segmentation neural network estimates, based on its training, that the given pixel illustrates a substance spill. The presence of a substance spill on the solid surface is detected based on the determined spill classification values of the pixels of the first image.