Neural Network Image Segmentation for Object Detection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting abnormalities inside animal or plant objects using images are inefficient and inaccurate due to reliance on manual analysis.
Innovation Solution
An image recognition method utilizing a pre-trained neural network model to segment images into regions and identify target objects within bounding boxes, trained with positive and negative samples to differentiate between target objects and noise, thereby improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual detection method is used to identify target objects in images, then the system complexity is low, but the detection accuracy is insufficient and information distortion occurs
Solution Approach 1:
The patent replaces the manual mechanical detection system with an automated neural network-based image recognition system. The target detection model automatically identifies target objects in images without human intervention, substituting the mechanical manual analysis process with an intelligent automated system that uses deep learning algorithms to achieve higher detection accuracy while reducing information distortion.
Solution Approach 2:
The patent introduces a target detection model as an intermediary between the input image and the final detection result. This neural network model acts as a mediator that processes the image data, extracts features, and identifies target objects, thereby improving detection accuracy while managing system complexity through modular architecture.
2Reliability
If noise filtering is performed on the image, then the noise is reduced, but information distortion occurs and target object recognition accuracy decreases
Solution Approach 1:
The patent applies preliminary action by training the neural network model in advance with both clean images and noisy images containing various types of noise. This pre-training enables the model to learn robust feature representations that are invariant to noise, allowing it to accurately recognize target objects without requiring post-processing noise filtering that could cause information distortion.
Solution Approach 2:
The patent converts the harmful effect of noise into a beneficial training condition. By intentionally adding various types of noise to training images and using them to train the neural network, the model learns to tolerate and even leverage noisy conditions, transforming what would normally be a detrimental factor into a means of improving the model's robustness and generalization capability.
Data Source
AI summary
Image recognition may include obtaining a first image, segmenting the first image into a plurality of first regions by using a target model, and searching for a target region among bounding boxes in the first image that use points in the first regions as centers. The target region is a bounding box in the first image in which a target object is located. The target model is a pre-trained neural network model configured to recognize from an image, a region in which the target object is located. The target model is obtained through training by using positive samples with a region in which the target object is located marked and negative samples with a region in which a noise is located marked. The target region is marked in the first image to improve accuracy of target object detection in an image.