A High Precision Counting Framework for Cerithidea moerchii towards Low Power Implementation
Hang Zhang, Takafumi Katayama, Tian Song, Takashi Shimamoto, Naotomo OTA · 2023
In this work, an adaptive framework which combines traditional image processing approach and machine learning based object detection approach to achieve low power implementation for the counting of Cerithidea moerchii (C. Moerchii) is proposed. The proposed framework categorizes the observing conditions into four types according to different environments of C. moerchii and adaptively select the counting tools with optimized computation complexity and detection accuracy. This proposal is characterized by the ability to distinguish different environments, switch counting modes, and save computational complexity with acceptable accuracy. The simulation results show that totally over 80% counting accuracy can be achieved by the proposed framework.