# Enhancing Landmark Point Detection in Eriocheir Sinensis Carapace with Differentiable End-to-End Networks

## TL;DR

This paper introduces a neural network that accurately detects 37 key points on the carapace of Chinese mitten crabs to improve aquaculture efficiency.

## Contribution

The study proposes a DSNT-based CNN model that outperforms others in accuracy and power efficiency for crab carapace landmark detection.

## Key findings

- The DSNT-based CNN achieved the highest R2 score of 0.9906 for landmark detection.
- The model showed superior computational efficiency and lower power consumption compared to alternatives.
- The framework reduces manual labor in crab breeding and quality inspection.

## Abstract

This study investigates three convolutional neural network (CNN) architectures to detect 37 key points on the carapace of the Chinese mitten crab. To enhance model generalization, the dataset was augmented with random distortions, rotations, and occlusions. The models were evaluated based on their detection accuracy, generalization capability, and power consumption. The results indicate that the network incorporating the DSNT module outperformed others in both accuracy and resource efficiency. These findings highlight the potential of the DSNT-based network to significantly enhance the efficiency and precision of quality assessment and monitoring in Chinese mitten crab aquaculture.

This research proposes using a neural network to detect and identify the landmark points of the carapace of the Chinese mitten crab, with the aim of improving efficiency in observation, measurement, and statistics in breeding and sales. A 37-point localization framework was developed for the carapace, with the dataset augmented through random distortions, rotations, and occlusions to enhance generalization capability. Three types of convolutional neural network models were used to compare detection accuracy, generalization ability, and model power consumption, with different loss functions compared. The results showed that the Convolutional Neural Network (CNN) model based on the Differentiable Spatial to Numerical Transform (DSNT) module had the highest R2 value of 0.9906 on the test set, followed by the CNN model based on the Gaussian heatmap at 0.9846. The DSNT-based CNN model exhibited optimal computational efficiency, particularly in power consumption metrics. This research demonstrates that the CNN model based on the DSNT module has great potential in detecting landmark points for the Chinese mitten crab, reducing manual workload in breeding observation and quality inspection, and improving efficiency.