Restricted-Zone Image Set Synthesis for Anomalous Organism Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing deep learning networks for detecting anomalous organisms in restricted zones face accuracy issues due to the lack of anomalous organisms in training data and inconsistencies in synthesized images, leading to obtrusive and unrealistic target images.

Innovation Solution

A method for generating an image set by segmenting target images from a first image set and synthesizing them with background images of restricted zones, using instance and semantic segmentation techniques, along with dimension and color adjustments, to create authentic sample images for improved detection accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If target images are synthesized with background images to create training data, then detection capability for anomalous organisms is improved, but the synthesized images become obtrusive and unrealistic

Engineering Contradiction:

Improvedetection capabilityVSAvoidimage realism

Core Design Contradiction:

ReliabilityVSManufacturing precision

Solution Approach 1:

The patent applies parameter changes by adjusting the dimension parameters of target images before synthesis. Specifically, the target image is resized to a predetermined dimension that matches the expected size of the anomalous organism in the restricted zone, ensuring proportional accuracy and realism in the synthesized image

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent segments the target image from the first image set using instance segmentation technology, separating the anomalous organism from the background. This segmentation allows for precise extraction and subsequent synthesis with the background image, maintaining image realism while preserving detection capability

Inventive Principle:

Principle #1Segmentation

2Ease of manufacture

If existing deep learning networks are trained without anomalous organisms in training data, then model training is simplified, but detection accuracy deteriorates

Engineering Contradiction:

Improvetraining simplicityVSAvoiddetection accuracy

Core Design Contradiction:

Ease of manufactureVSMeasurement precision

Solution Approach 1:

The patent creates synthetic training data by copying and combining elements from existing images. Target images are extracted from the first image set and synthesized with background images from the restricted zone, creating realistic training samples that improve detection accuracy without requiring actual anomalous organisms in the training data

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent performs preliminary actions by pre-processing the first image set to extract target images and pre-processing background images from the restricted zone. These pre-processed images are then combined to create the training data set, enabling accurate detection model training before actual detection occurs

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20250245785A1Method and device for generating image set, and computer-readable storage medium

Publication Date: 2025.07.31 BOE TECHNOLOGY GROUP CO LTD

AI summary

Embodiments of the present disclosure provide an image set generation method. The image set is configured to train a detection model for an anomalous organism in a restricted zone and comprises a plurality of sample images; and the method comprises generating each sample image according to steps as below: acquiring at least one target image according to a pre-acquired first image set, wherein the target image is an image of the anomalous organism that is segmented from a first image in the first image set; and synthesizing the target image with a background image to acquire the sample image, wherein the first image set comprises a plurality of first images, and the background image is acquired by shooting the restricted zone.