Marine Life Classification Using Deep Neural Network Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning algorithms struggle to reliably distinguish multiple instances of marine life objects in a single image, necessitating more sophisticated image processing capabilities for effective underwater organism detection and classification.
Innovation Solution
A Deep Neural Network (DNN) system with subnetworks, including a region proposal network, classification network, and semantic segmentation network, is employed to identify and classify marine life objects by generating feature maps, determining target object probabilities, calculating image coordinates, and applying masking layers for instance segmentation and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional sonar images are used for detection and recognition of underwater organisms, then the system is simpler, but the detection precision and classification accuracy are insufficient
Solution Approach 1:
The system segments the image processing task into multiple specialized subnetworks: a backbone network for feature extraction, a region proposal network for candidate generation, a classification network for species identification, and a segmentation network for instance separation. Each subnetwork focuses on a specific aspect of the detection pipeline, improving overall precision while managing complexity through functional decomposition.
Solution Approach 2:
The system transitions from traditional 2D sonar imaging to multi-dimensional deep learning feature spaces. By extracting hierarchical features through multiple convolutional layers and processing them through specialized networks, the system creates additional dimensional representations that enhance detection precision beyond what traditional sonar methods provide.
2Reliability
If simple deep learning algorithms are used, then the system is easier to implement, but the ability to distinguish multiple instances of marine life objects in a single image is insufficient
Solution Approach 1:
The system employs instance segmentation technology that separates multiple marine life objects within a single image into distinct instances. The segmentation network generates unique masks for each detected object, enabling reliable distinction between multiple instances of the same species or different species overlapping in the same image frame.
Solution Approach 2:
The region proposal network acts as an intermediary between the backbone feature extractor and the final classification/segmentation networks. It generates candidate region proposals that are then processed by subsequent networks, mediating the complex task of multi-instance detection by breaking it down into manageable candidate evaluation steps.
3Measurement precision
If sophisticated deep learning algorithms with multiple subnetworks are used, then the classification and segmentation accuracy improve, but the training time and computational resources increase
Solution Approach 1:
By segmenting the neural network into modular subnetworks (backbone, region proposal, classification, segmentation), the system enables independent training and optimization of each component. This modular approach allows for more efficient training compared to training a monolithic network, as each subnetwork can be trained with focused loss functions and updated independently.
Solution Approach 2:
The backbone network performs preliminary feature extraction that is reused across multiple subsequent networks. By pre-computing rich feature representations in the backbone, the system avoids redundant computation in later stages, reducing overall training time while maintaining high classification and segmentation accuracy.
Data Source
AI summary
The present disclosure provides a method and system architectures for carrying out the method of automated marine life object classification and identification utilising a core of a Deep Neural Network, DNN, to facilitate the operations of a post-processing module subnetwork such as instance segmentation, masking, labelling, and image overlay of an input image determined to contain one or more target marine life objects. Multiple instances of target objects from the same image data can be easily classified and labelled for post-processing through application of a masking layer over each respective object by a semantic segmentation network.