Juvenile Fish Limb Segmentation With Multi-Scale Cascaded CNN
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for fish phenotype identification, particularly for juvenile fish, are inefficient and inaccurate due to the small number of pixels in their limbs, making it difficult to classify these pixels effectively.
Innovation Solution
A juvenile fish limb identification method based on a multi-scale cascaded perceptual convolutional neural network that includes image preprocessing, semantic annotation, and a multi-scale cascaded perceptual convolutional neural network for feature extraction, candidate region generation, and fish limb mask generation, utilizing an Attention-RPN for improved feature mapping and noise removal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional identification networks are used for juvenile fish limbs, then the network structure is simple, but the classification accuracy is low due to small number of pixels in limbs
Solution Approach 1:
The network is divided into multiple cascaded stages (first identification network, second identification network, third identification network), where each stage progressively refines the classification of limb pixels. This segmentation allows the system to handle small pixel counts by breaking down the classification task into manageable steps, improving accuracy without requiring an overly complex single-stage network.
Solution Approach 2:
The patent introduces multi-scale feature extraction by processing images at different resolutions and scales. This dimensional approach allows the network to capture limb features at various levels of detail, effectively compensating for the small number of pixels by utilizing spatial information from multiple scales simultaneously.
2Measurement precision
If multi-scale cascaded perceptual convolutional neural network is used, then classification accuracy is improved, but computational complexity increases
Solution Approach 1:
The first identification network performs preliminary classification of limb pixels before subsequent networks process the data. This preliminary action filters and pre-processes the input, reducing the computational burden on later stages while maintaining high accuracy through progressive refinement.
Solution Approach 2:
The cascaded network structure dynamically adjusts processing based on confidence levels and feature prominence. Each stage adapts its processing based on the output of previous stages, allowing the system to focus computational resources where they are most needed rather than uniformly processing all data at maximum complexity.
3Measurement precision
If traditional methods are used for phenotype identification, then the method is simple, but it cannot effectively classify small-scale limb pixels of juvenile fish
Solution Approach 1:
The network applies different processing strategies to different regions of the image, with specialized attention to limb regions. Local quality enhancement techniques are applied specifically to small-scale limb pixels, allowing effective classification of these critical regions without unnecessarily complicating the processing of other image areas.
Solution Approach 2:
Multi-scale feature extraction processes images at different resolutions, creating additional dimensional information about limb structures. This dimensional approach enables effective classification of small-scale pixels by viewing them through multiple spatial lenses, capturing features that would be invisible at a single scale.
4Object-affected harmful factors
If non-contact acquisition method is used, then stress and physical damage to fish are reduced, but image processing complexity increases
Solution Approach 1:
Image preprocessing steps including normalization, noise filtering, and feature enhancement are performed before main classification. This preliminary action prepares the images for more efficient processing in subsequent stages, reducing the overall computational complexity while maintaining the benefits of non-contact acquisition.
Solution Approach 2:
The processing pipeline dynamically adjusts based on image quality and fish movement. Adaptive processing techniques optimize computational resources based on the specific characteristics of each captured image, reducing unnecessary processing complexity while maintaining high classification accuracy for limb identification.
Data Source
AI summary
The present disclosure provides a juvenile fish limb identification method based on a multi-scale cascaded perceptual convolutional neural network. The method includes the following steps: acquiring a video sequence of a juvenile fish, dividing a fish body into five non-overlapping parts, performing semantic annotation on the five non-overlapping parts, and taking the five non-overlapping parts as an input of the multi-scale cascaded perceptual convolutional neural network; and using a convolutional layer as a feature extractor, performing feature extraction on an input image containing the annotation of each limb, inputting extracted features into an Attention-region proposal network (RPN) structure, determining a category of each pixel, and generating a limb mask of each limb category using a multi-scale cascade method. According to the method, the limbs of the juvenile fish can be efficiently and accurately identified, and technical support is provided for posture quantification of the juvenile fish.