Maritime Obstacle Detection with Adaptive Image Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to effectively acquire object information from images captured in environments with environmental noise, such as fog, using artificial neural networks, and there is a need to improve training efficiency of these networks.
Innovation Solution
A method utilizing a first artificial neural network to process images with low noise levels and a second neural network to reduce noise in high-noise images, combined with pre-processing techniques like RGB normalization and data augmentation, to enhance image segmentation and object information acquisition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If artificial neural networks are used to acquire object information from images with environmental noise, then object recognition capability is improved, but training efficiency deteriorates
Solution Approach 1:
The patent applies preliminary action by performing pre-processing on training images before feeding them to the neural network. This includes adding synthetic environmental noise (fog, rain, snow) to clean images during the training phase, so the network learns to recognize objects under various noisy conditions without requiring extensive real-world noisy training data, thereby improving training efficiency while maintaining recognition accuracy
2Manufacturing precision
If noise reduction processing is applied to images with environmental noise, then image quality is improved, but processing complexity increases
Solution Approach 1:
The patent uses an intermediary approach by introducing a dedicated noise reduction module that processes images before they enter the main object recognition neural network. This separate module specifically handles environmental noise (fog, rain, snow) using specialized algorithms, allowing the main network to focus on object recognition without being burdened by noise processing complexity
3Measurement precision
If multiple processing steps are used to handle environmental noise, then object information acquisition accuracy is improved, but system complexity increases
Solution Approach 1:
The patent merges multiple processing functions into an integrated system architecture. The noise reduction module, pre-processing unit, and object recognition neural network are combined into a unified processing pipeline that handles environmental noise and object detection in a coordinated manner, reducing overall system complexity while maintaining high accuracy through the synergistic effect of multiple processing steps
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to a method for acquiring object information, the method comprising: a step of obtaining an input image capturing a sea; a step of obtaining a noise level of the input image; a first image processing step of when the noise level indicates a noise lower than a predetermined level, acquiring object information related to an obstacle included in the input image from the input image by using a first artificial neural network wherein the first artificial neural network outputs, from an image, object information related to at least one of a distance and a type of an object included in the image; and a second image processing step of when the noise level indicates a noise higher than the predetermined level, obtaining an image of which an environmental noise is reduced from the input image by using a second artificial neural network trained to generate, from a first image containing a noise, a second image of which a noise is reduced, and acquiring object information related to an obstacle included in the sea from the image of which the environmental noise is reduced by using the first artificial neural network, wherein the second artificial neural network is trained by using a first error calculated from a third artificial neural network discriminating the second image and an image for comparison corresponding to the first image and having a noise lower than the first image.