Design and Experimental Evaluation of a Bottom-Drain and Modified-Screw Lifting System for Efficient Adult Fish Harvesting and Digital Sorting in Industrialized Aquaculture Systems

To address the challenges of low efficiency, high labour intensity, severe fish damage, and insufficient intelligence in adult fish harvesting within industrialised aquaculture systems, this study developed an integrated harvesting and sorting system incorporating bottom-drain fish collection, automatic herding, a modified screw-lifting mechanism, and digital grading. The system employs a bottom-discharge port on the culture tank—controlled by an electric ball valve—to replace manual net-herding, enabling simultaneous fish-and-water drainage. An automatic herding cart with a retractable net (equipped with surface floats and bottom weights) achieves bottom-hugging herding to prevent fish escape. A multi-column convolutional neural network (MCNN) fish-density recognition model was deployed above the collection tank to adaptively adjust the push-board position, thereby avoiding compression injury. The screw blade head was modified by removing the inlet-edge portions to eliminate the fish-cutting zone, reducing the mean damage rate from 1.8% to 1.32% (N = 12; 66.7% of trials recorded zero damage). Twelve repeated full-scale harvesting trials were conducted using largemouth bass (Micropterus salmoides; body length 20–30 cm, body mass 300–500 g) in a 2.5 m diameter PP circular tank. Results showed a mean collection efficiency of 95.94 ± 3.33% (p < 0.001), a mean damage rate of 1.32 ± 1.96%, and a mean sorting accuracy of 96.83 ± 2.50% (p = 0.014). The MCNN model attained a recognition accuracy of 78–85%. The proposed system realises automated, intelligent, and low-damage fish harvesting, providing technical support for fully automated production-line aquaculture.