隨著智慧養殖技術的發展，如何以低成本方式獲取魚隻位置資訊，成為影響水質監控與行為分析準確性的關鍵。本研究提出一套整合深度學習與自適應校正機制的單目魚隻距離測量方法，採由上往下拍攝，搭配軟體演算法補償單目視覺的限制，以取代高成本的雙目攝影設備。研究以三種觀賞魚（皇冠六間、孔雀魚、斑馬魚）為對象，進行三項實驗：比較不同影像前處理方式（原始圖、灰度圖、邊緣圖）對模型效能的影響；評估資料增生策略（幾何、擾動、混合）對偵測表現的改善幅度；並透過試誤法找出不同魚種的最佳自適應距離校正參數。
實驗採用 YOLOv8s、YOLOv8m、Faster R-CNN、Mask R-CNN 等四種模型進行分析，結果顯示 YOLOv8m 整體表現最優，於多數資料組合下 mAP50 達 0.98 以上，mAP75 可達 0.94。影像前處理方面，以原始圖搭配 YOLOv8m 可取得最高準確率，而邊緣圖整體表現偏低。資料增生策略中，以混合式增生效果最佳，平均提升mAP50 約 6%，尤其對灰度圖效果顯著。自適應校正方面，提出之補償公式能根據魚隻長寬比自動調整距離估算值，在測試集中平均可降低誤差約 12%～20%。綜合而言，本研究在小型物件偵測、資料前處理與距離估測領域皆展現出高度實用性與擴充潛力，未來可應用於智慧養殖、水族箱監控等情境，具顯著研究與實務價值。

With the rapid development of intelligent aquaculture technologies, achieving effective and low-cost fish position monitoring has become a crucial task. In particular, during water quality monitoring or fish behavior analysis, accurate fish location information significantly affects sensor data precision and farming decisions. This study aims to develop a monocular fish distance measurement system that integrates deep learning and adaptive correction. By adopting a top-down camera perspective and software-based compensation for hardware limitations, the system reduces the need for expensive equipment such as stereo cameras.
We employed four object detection models—YOLOv8 (s and m versions), Faster RCNN, and Mask R-CNN—for fish detection. The research is divided into three experimental parts: (1) image preprocessing, where small datasets were created using three common ornamental fish species—Frontosa (Cyphotilapia frontosa), Guppy (Poecilia reticulata), and Zebrafish (Danio rerio)—to compare the detection performance of original, grayscale, and edge-processed images; (2) data augmentation strategies, including geometric transformations (e.g., flipping, rotation), image perturbations (e.g., blur, noise), and their combinations to assess improvements in modelaccuracy; and (3) optimization of custom correction formula parameters through iterative testing to enhance distance estimation precision for each species.
Experimental results show that in small fish detection tasks, original and grayscale images outperform edge images in terms of mAP, suggesting that small object detection still relies heavily on color and texture features. In terms of data augmentation, combining geometric and perturbation-based techniques significantly boosts model performance. Notably, YOLOv8m achieved near-perfect accuracy in both mAP50 and mAP75. Finally, the proposed adaptive correction formula effectively adjusts estimated distance based on differences between the detected bounding box length and the ideal ratio. The method adapts well to different species' aspect ratio variations, improving both the practicality and generalizability of monocular distance measurement.
In conclusion, this research successfully constructs a low-cost, high-accuracy, and highly scalable fish distance measurement system with promising application potential. It is expected to become an essential component in smart aquaculture observation systems.

[1]. Bianchi, M. C. G., Chopin, F., Farme, T., Franz, N., Fuentevilla, C., Garibaldi, L., & Laurenti, A. L. G. (2014). FAO: the state of world fisheries and aquaculture. Food and Agriculture Organization of the United Nations: Rome, Italy, 1-230.
[2]. Food and Agriculture Organization of the United Nations. (2024). The state of world fisheries and aquaculture 2024 (SOFIA 2024). https://www.fao.org/publications/sofia/en/
[3]. Engle, C., Kumar, G., & van Senten, J. (2019). Intelligent aquaculture. World Aquaculture Society. Retrieved from https://www.was.org/article/Intelligent-aquaculture.aspx
[4]. Huang, Y. P., & Khabusi, S. P. (2025). Artificial Intelligence of Things (AIoT) Advances in Aquaculture: A Review. Processes, 13(1), 73.
[5]. Cui, M., Liu, X., Liu, H., Zhao, J., Li, D., & Wang, W. (2024). Fish Tracking, Counting, and Behaviour Analysis in Digital Aquaculture: A Comprehensive Review. arXiv preprint arXiv:2406.17800.
[6]. Zhao, Y., Qin, H., Xu, L., Yu, H., & Chen, Y. (2025). A review of deep learning-based stereo vision techniques for phenotype feature and behavioral analysis of fish in aquaculture. Artificial Intelligence Review, 58(1), 1-61.
[7]. Kunapinun, A., Fairman, W., Wills, P. S., Mejri, S., Kostelnik, M., & Ouyang, B. (2024, June). Innovative aquaculture biometrics analysis: harnessing IR lasers and ToF cameras for microscopic fish larvae tracking. In Ocean Sensing and Monitoring XVI (Vol. 13061, pp. 113-120). SPIE.
[8]. Sterzelecki, F. C., dos Santos Cipriano, F., Vasconcelos, V. R., Sugai, J. K., Mattos, J. J., Derner, R. B., ... & Cerqueira, V. R. (2021). Minimum rotifer density for best growth, survival and nutritional status of Brazilian sardine larvae, Sardinella brasiliensis. Aquaculture, 534, 736264.
[9]. Xiao, D., Ma, Q., Xie, Y., Zheng, Q., & Zhang, Z. (2018). A power-frequency electric field sensor for portable measurement. Sensors, 18(4), 1053.
[10]. Tallon, B., Roux, P., Matte, G., Guillard, J., & Skipetrov, S. E. (2020). Acoustic density estimation of dense fish shoals. The Journal of the Acoustical Society of America, 148(3), EL234-EL239.
[11]. Terayama, K., Shin, K., Mizuno, K., & Tsuda, K. (2019). Integration of sonar and optical camera images using deep neural network for fish monitoring. Aquacultural Engineering, 86, 102000.
[12]. Liu, H., Ma, X., Yu, Y., Wang, L., & Hao, L. (2023). Application of deep learning-based object detection techniques in fish aquaculture: a review. Journal of Marine Science and Engineering, 11(4), 867.
[13]. Mei, J., Hwang, J. N., Romain, S., Rose, C., Moore, B., & Magrane, K. (2021, June). Absolute 3d pose estimation and length measurement of severely deformed fish from monocular videos in longline fishing. In ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 2175-2179). IEEE.
[14]. Chen, Z., Zhang, Z., Dai, F., Bu, Y., & Wang, H. (2017). Monocular vision-based underwater object detection. Sensors, 17(8), 1784.
[15]. Xu, Z., Wang, X., & Deng, Y. (2020). Rotating focused field eddy-current sensing for arbitrary orientation defects detection in carbon steel. Sensors, 20(8), 2345.
[16]. Grimaldi, M., Nakath, D., She, M., & Köser, K. (2023). Investigation of the challenges of underwater-visual-monocular-slam. arXiv preprint arXiv:2306.08738.
[17]. Fu, C., Liu, R., Fan, X., Chen, P., Fu, H., Yuan, W., ... & Luo, Z. (2023). Rethinking general underwater object detection: Datasets, challenges, and solutions. Neurocomputing, 517, 243-256.
[18]. Zhang, X., Wang, Z., Liu, D., Lin, Q., & Ling, Q. (2020). Deep adversarial data augmentation for extremely low data regimes. IEEE Transactions on Circuits and Systems for Video Technology, 31(1), 15-28.
[19]. Park, K., Chae, M., & Cho, J. H. (2021). Image pre-processing method of machine learning for edge detection with image signal processor enhancement. Micromachines, 12(1), 73.
[20]. Egger, J., Gsaxner, C., Pepe, A., Pomykala, K. L., Jonske, F., Kurz, M., ... & Kleesiek, J. (2022). Medical deep learning—A systematic meta-review. Computer methods and programs in biomedicine, 221, 106874.
[21]. Ferentinos, K. P. (2018). Deep learning models for plant disease detection and diagnosis. Computers and electronics in agriculture, 145, 311-318.
[22]. Shorten, C., & Khoshgoftaar, T. M. (2019). A survey on image data augmentation for deep learning. Journal of big data, 6(1), 1-48.
[23]. Liang, W., Liang, Y., & Jia, J. (2023). MiAMix: enhancing image classification through a multi-stage augmented mixed sample data augmentation method. Processes, 11(12), 3284.
[24]. Zheng, L., Tse, T. H. E., Wang, C., Sun, Y., Chen, H., Leonardis, A., ... & Chang, H. J. (2024). GeoReF: Geometric Alignment Across Shape Variation for Category-level Object Pose Refinement. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 10693-10703).
[25]. Kaur, P., Khehra, B. S., & Mavi, E. B. S. (2021, August). Data augmentation for object detection: A review. In 2021 IEEE International Midwest Symposium on Circuits and Systems (MWSCAS) (pp. 537-543). IEEE.
[26]. Liu, X., Cheng, M., Zhang, H., & Hsieh, C. J. (2018). Towards robust neural networks via random self-ensemble. In Proceedings of the european conference on computer vision (ECCV) (pp. 369-385).
[27]. Li, T., Gang, Y., Li, S., & Shang, Y. (2025). A small underwater object detection model with enhanced feature extraction and fusion. Scientific Reports, 15(1), 2396.
[28]. Feng, C., Wang, C., Zhang, D., Kou, R., & Fu, Q. (2024). Enhancing Dense Small Object Detection in UAV Images Based on Hybrid Transformer. Computers, Materials & Continua, 78(3).
[29]. Bosquet, B., Cores, D., Seidenari, L., Brea, V. M., Mucientes, M., & Del Bimbo, A. (2023). A full data augmentation pipeline for small object detection based on generative adversarial networks. Pattern Recognition, 133, 108998.
[30]. Liu, L., Ouyang, W., Wang, X., Fieguth, P., Chen, J., Liu, X., & Pietikäinen, M. (2020). Deep learning for generic object detection: A survey. International journal of computer vision, 128, 261-318.
[31]. Kaur, R., & Singh, S. (2023). A comprehensive review of object detection with deep learning. Digital Signal Processing, 132, 103812.
[32]. Ultralytics. (2023). YOLOv8: The latest version of You Only Look Once. Retrieved May 2025, from https://ultralytics.com/yolov8
[33]. Tan, M., & Le, Q. (2019, May). Efficientnet: Rethinking model scaling for convolutional neural networks. In International conference on machine learning (pp. 6105-6114). PMLR.
[34]. Jocher, G., et al. (2023). YOLOv8: The latest evolution of the YOLO object detection series [Computer software]. Ultralytics. https://github.com/ultralytics/ultralytics
[35]. Ren, S., He, K., Girshick, R., & Sun, J. (2015). Faster r-cnn: Towards real-time object detection with region proposal networks. Advances in neural information processing systems, 28.
[36]. He, K., Gkioxari, G., Dollár, P., & Girshick, R. (2017). Mask r-cnn. In Proceedings of the IEEE international conference on computer vision (pp. 2961-2969).
[37]. Panigrahi, S., Nanda, A., & Swarnkar, T. (2020). A survey on transfer learning. In Intelligent and Cloud Computing: Proceedings of ICICC 2019, Volume 1 (pp. 781-789). Singapore: Springer Singapore.
[38]. Wang, Y., Yao, Q., Kwok, J. T., & Ni, L. M. (2020). Generalizing from a few examples: A survey on few-shot learning. ACM computing surveys (csur), 53(3), 1-34.
[39]. Kisantal, M., Wojna, Z., Murawski, J., Naruniec, J., & Cho, K. (2019). Augmentation for small object detection. arXiv preprint arXiv:1902.07296.
[40]. Keshari, R., Vatsa, M., Singh, R., & Noore, A. (2018). Learning structure and strength of CNN filters for small sample size training. In proceedings of the IEEE conference on computer vision and pattern recognition (pp. 9349-9358).