Cephalopods Classification Using Fine Tuned Lightweight Transfer Learning Models

Cephalopods identification is a formidable task that involves hand inspection and close observation by malacologist. Manual take time are always contingent on the involvement of experts. A system proposed to alleviate this challenge uses transfer learning techniques classify cephalopods automatically. In method, only Lightweight pre-trained networks chosen enable IoT in cephalopod recognition. First, efficiency models determined evaluating their performance comparing findings. Second, fine-tuned adding dense layers tweaking hyperparameters improve classification accuracy. The also employ well-tuned Rectified Adam optimizer increase accuracy rates. Third, with Gradient Centralisation (RAdamGC) used reduce training time. framework enables an Internet Things (IoT) or embedded device perform tasks embedding suitable lightweight network. models, MobileNetV2, InceptionV3, NASNet Mobile have achieved 89.74%, 87.12%, respectively. findings indicated can different kinds cephalopods. results demonstrated there significant reduction RAdamGC.