The Nofima team remains optimistic that the boundaries of optical technology can be pushed further

Advanced Camera Scanning Holds Promise for Aquaculture but Fails to Sort Juvenile Atlantic Cod by Sex

NORWAY
Thursday, May 21, 2026, 05:00 (GMT + 9)

High-tech non‑invasive imaging methods struggle with biological limitations in early growth stages forcing researchers back to the drawing board for sustainable fish farming solutions

A groundbreaking research initiative aiming to revolutionize the aquaculture industry has hit a biological roadblock. Scientists from the food research institute Nofima have announced that advanced camera measurements and imaging technologies cannot currently be used to sort farmed Atlantic cod by sex at an early stage without physical intervention.

Click on the image to enlarge it

The CodSex project, funded by FHF ( the Norwegian Seafood Research Fund ), was launched to find a quick, gentle, and non‑invasive way to distinguish male and female fish. Today's safe sorting method requires dissection—cutting into dead fish to see their reproductive organs, known as gonads—or using ultrasound, which lacks accuracy in young fish. Cod farmers are eager to separate the sexes early in production to optimize selective breeding, improve broodstock management, and enhance production planning to reduce resource use.

The Limits of Light Waves

Led by Nofima scientist Samuel Ortega , the project tested the Maritech Eye hardware platform under industrial conditions to see if light waves penetrating the skin could reveal a fish's sex. The primary technology explored was hyperspectral imaging, an advanced form of spectroscopy that analyzes how light interacts with biological tissue to see details invisible to the naked eye.

Ultimately, the computer models could not reliably distinguish the sexes because color signals from the whole fish were too weak and unclear.

"Unfortunately, none of the techniques investigated currently provide a robust solution for early and non‑invasive sex determination in Atlantic cod under industrial conditions," concludes Ortega. "The industry wants a tool that can sort fish quickly and gently without physical intervention, but we must acknowledge that the technology currently faces optical and biological limitations on young fish."

Promising Clues Hidden in the Gonads

While scanning the exterior of whole juvenile fish failed, researchers found success when looking deeper. By utilizing LDA ( Linear Discriminant Analysis )—a statistical machine learning method—scientists examined measurements taken directly from the gonads.

The computers could clearly distinguish males from females based on how the reproductive tissue reflected light. When this data-driven model was applied to whole fish, it captured sex‑related information more consistently across different fish sizes. However, there is a catch: this only works on larger, more mature cod that already have well‑developed gonads. Because the differences are too subtle in younger fish, the technology cannot yet help farmers at the fry stage.

Representation of hyperspectral fluorescence images acquisition 

Four Rigorous Trials

The conclusions were reached after Nofima conducted four extensive trials on numerous intact, dead cod at various developmental stages, closely collaborating with industry partners Vesterålen Havbruk and Kime Akva :

• Trial 1: Researchers examined 188 small fish from Vesterålen Havbruk with undeveloped gonads. Ventral and lateral hyperspectral images were taken, but ultrasound proved completely ineffective on such small individuals.

• Trial 2: A group of 136 larger fish from Kime Akva was analyzed. Ultrasound showed a low accuracy of around 40 percent and was deemed unreliable, requiring dissection for sex verification.

• Trial 3: A detailed trial of 30 fish from Nofima’s cod breeding program utilized hyperspectral imaging, direct gonad measurements, and fluorescence imaging, which highlights how light is emitted from illuminated tissue.

• Trial 4: A massive sample of 998 fish from Nofima’s breeding program was photographed using standard RGB (red, green, and blue) color images, with a subset also undergoing ventral hyperspectral scanning.

Example of RGB images used for CNN classification

The researchers also tested whether the fish's surface featured sex-specific light emission (known as biofluorescence) and used artificial intelligence to search for external differences in standard color photos. Both of these experiments failed to yield reliable results.

A Tech Revolution with Future Potential

Despite the current setback for young cod, non-invasive spectroscopy has accelerated rapidly in recent years and is already transforming industrial food processing conveyor belts at high speeds.

The technology is actively used in commercial settings to determine melanin and blood spots in salmon, measure the fat content in salmon, mackerel, and herring, detect defects in fish fillets, assess water content in stockfish and dried salted cod, evaluate quality in chicken fillets, and even measure the meat content in crab legs—an industry worth millions of US$ .

Nofima scientist Samuel Ortega is leading efforts to test imaging technology for determining the sex of farmed cod as early as possible in its life stage. Photo: Ronald Johansen, Nofima

The Nofima team remains optimistic that the boundaries of optical technology can be pushed further.

"Determining sex at early and intermediate growth stages is important for what we call selective breeding. Earlier and more accurate sex identification can provide better broodstock management, more efficient breeding and improved production planning, thereby reduce resource use and contribute to increased sustainability in cod farming," says Ortega. "Additionally, although fluorescence imaging and conventional imaging did not produce reliable results in this study, these approaches may still hold greater potential for future improvements in image‑based sex identification."

🇯🇵 Japanese

Highly promising aquaculture industry, gender selection of juvenile Atlantic fish, failed.

The non-invasive イメージング technique of ハイテクな, the initial growth stage, the limits of biology, and the research The researcher is a possible solution to the problem of なタラ breeding.

ノルウェー—The revolution of the breeding industry has begun, the research on the period of the painting, and the obstacles of biology are on the rise. Food Research Institute Nofima 's scientists, height measurement technology, current technology, body's intervention technology, breeding technology, Atlantic OceanをIn the early stage, the gender is selected and the gender is selected.

FHF Funding provided by (ノルウェーFishery Products Research Fund )したCodSexプロジェクトは, fish injuryつけずにquicklyかつnon-invasiveにmaleとfemaleをidentificationするmethodを见つけるために立ち上げられた. The current method of selecting and distinguishing is the anatomy If the accuracy of the fish is insufficient, the ultrasonic examination will be checked using the するしかない. Agriculturists, systematic selection and breeding, optimization, broodstock management, improvement, production planning, strengthening, consumption of resources There is a high possibility of maintaining the breeding and breeding, and the early stages of production are divided into genders, and the hope is high.

The limits of light waves

Nofima 's scientific scholar , Maritech under industrial conditions Eye protection, skin clear Over the light wave が鱼の性を明らかにできるかどうかを検证した. Investigate the technology of the master, the technology of the technology, and the biological tissue of the light The interaction is analyzed by the naked eye and the details are confirmed by the naked eye, and the spectroscopic method is a kind of method.

The final に, the whole fish is weak and the fish is not clear. .

"Senami Nako, Investigating Nako's technology, Nako's early non-intrusive gender determination under industrial conditionsに対して、Now ロバストなsolved decisionをprovidesするものは高々ありません」とオルテガは Conclusionづけている. "Industry は, body な intervention な し に, rapid か つ し く 鱼 を 选 で き る ツ ー ル を Requirement め て い ま す が, now のTechnology is a fish, it is a fish, it is an optics, it is a biology, it is a limit, it is a direct face, it is a problem, it is a recognition, it is a problem.”

Gonads and hands

All the juvenile fish have failed, and the researchers have failed. The statistical technique of machine learning, LDA ( Linear Discriminant Analysis ), is used, and the scientist's reproductive glands are used to directly obtain the measurement certificate.

Reproductive tissue が光をどのように Reflection するかにbased づいて, コンピュータは Male とfemale を Definition することができた. All このデータ駆 type モデルをfish are suitable for all kinds of したところ and different なる鱼のサイズにわたって, より一 consistent してgender-related information の杀えることができた.しかし, これには下とし点がある.これは, gonad がすでに十に発达した, より大きく mature したタラにしかfunction しないのである. If there is a difference between the fish and fish, the fish is fine and the fish is fine, and the fish breeder is a fish breeder at the juvenile stage.

4つの即格なtrial

この結論は、Nofima が、業界パートナーである Vesterålen Havbruk および Kime Akva と密接に協力し、さまざまな成長・開発段階にある多数の無傷の（死んだ）タラに対して4つの広範な試験を実施した後に導き出された。

• 試験1: 生殖腺が目視で発達していない Vesterålen Havbruk 由来の小さな魚、計 188 匹を調査した。腹側および側面からハイパースペクトル画像が撮影されたが、超音波はこのような小さな個体にはまったく効果がないことが証明された。

• 試験2: Kime Akva 由来の、より大きな魚、計 136 匹を分析した。超音波の精度は約 40パーセント と低く、信頼できないと判断されたため、性別の判定には解剖が必要であった。

• 試験3: Nofima のタラ育種プログラムからの 30 匹の魚を対象に、より詳細な小規模試験が行われた。ハイパースペクトル画像に加えて、光を照射した後に組織から放出される光を示し、組織タイプ間の違いを際立たせることができる蛍光画像が撮影された。生殖腺から直接の測定も行われた。

• 試験4: Nofima の育種プログラムからの計 998 匹の魚が、標準的な RGB（赤、緑、青の光に基づく標準的なカラー画像）を使用して撮影され、その一部のサブセットでは腹側のハイパースペクトル画像も撮影された。

研究者たちはまた、魚の表面に性特異的な光放出（バイオフルオレッセンスと呼ばれる）があるかどうか、あるいは人工知能を用いたコンピュータモデルが普通の画像から魚の外見の違いを見出せるかどうかもテストした。これらの実験も結果を出すには至らなかった。

将来の可能性を秘めた技術革命

若いタラに対する現在の挫折にもかかわらず、非侵襲的な分光法は近年急速に加速しており、すでに工業生産のベルトコンベアに匹敵する速度で食品加工を変化させている。

この技術は、サーモンにおけるメラニンや血斑の判定、サーモン、サバ、ニシンなどの魚の脂肪分の測定、魚のフィレの欠陥検出、ストックフィッシュ（干物）や塩蔵乾燥タラの水分量の評価、チキンフィレの品質評価、さらにはカニの脚の肉含有量の測定など、何百万 USドル もの価値がある産業において、商業の現場で活発に使用されている。

Nofima のチームは、光学技術の限界をさらに押し広げることができると楽観視し続けている。

「初期および中間の成長段階で性別を判定することは、私たちが系統的選抜育種と呼ぶものにとって重要です。より早期でより正確な性別識別は、より良い親魚管理、より効率的な育種、そして改善された生産計画を提供することができ、それによって資源の消費を抑え、タラ養殖における持続可能性の向上に貢献します」とオルテガは語る。「さらに、今回の研究において蛍光イメージングや従来のイメージングは信頼できる結果を生み出しませんでしたが、これらのアプローチは、画像ベースの性別識別における将来の改善に向けて、依然としてより大きな可能性を秘めているかもしれません」

🇨🇳 Simplified Chinese (简体中文)

先进相机扫描为水产养殖带来希望但未能按性别筛选大西洋鳕鱼幼鱼

高科技非侵入性成像方法在早期生长阶段面临生物学局限，迫使研究人员回到原点以寻求可持续的鳕鱼养殖解决方案

挪威 — 一项旨在彻底改变水产养殖业的开创性研究计划遭遇了生物学障碍。来自食品研究所 Nofima 的科学家们宣布，目前的先进相机测量和成像技术无法在不进行物理干预的情况下，在早期阶段按性别筛选养殖的 大西洋鳕鱼。

由 FHF（挪威海鲜研究基金）资助的 CodSex 项目旨在寻找一种快速、温和且非侵入性的方法来区分雄鱼和雌鱼。如今，可靠的筛选方法需要进行解剖——即切开死鱼以查看被称为性腺的生殖器官——或者使用在幼鱼中缺乏足够准确度的超声波。鳕鱼养殖户渴望在生产早期将性别分开，以优化选择性育种、改善亲鱼管理并加强生产规划，从而减少资源消耗并有助于提高鳕鱼养殖的可持续性。

光波的局限性

该项目由 Nofima 科学家 萨缪尔·奥尔特加 领导，在工业条件下测试了 Maritech Eye 硬件平台，以观察穿透皮肤的光波是否能揭示鱼类的性别。所探索的核心技术是高光谱成像，这是一种基于光谱学的先进形式，通过分析光如何与生物组织相互作用来观察肉眼无法看到的细节。

最终，由于来自整条鱼的颜色信号过于微弱且不清晰，计算机模型无法可靠地区分性别。

“不幸的是，目前所研究的技术中，没有一种能够为工业条件下大西洋鳕鱼的早期和非侵入性性别鉴定提供鲁棒的解决方案，”奥尔特加总结道。“行业需要一种能够在没有物理干预的情况下快速且温和地筛选鱼类的工具，但我们必须承认，该技术目前在幼鱼上面临着光学和生物学的局限性。”

隐藏在性腺中的有望线索

虽然扫描整条幼鱼的外部失败了，但研究人员在探寻更深层部位时取得了成功。通过利用 LDA（线性判别分析）——一种用于机器学习的统计分类方法——科学家们对直接从性腺获取的测量数据进行了检验。

计算机能够根据生殖组织反射光线的方式，清晰地marshal区分雄性和雌性。当这种数据驱动的模型被应用于整条鱼时，它在不同鱼类个体大小和试验中捕获性别相关信息的内容表现得更加一致。然而，这其中有一个限制：这只适用于已经具有发育良好的性腺的较大、更成熟的鳕鱼。由于在较小的鱼中这种差异过于细微，该技术目前还无法帮助鱼苗阶段的养殖户。

四项严格的试验

These conclusions were reached after Nofima , in close collaboration with industry partners Vesterålen Havbruk and Kime Akva , conducted four extensive trials on large numbers of whole (dead) cod at different developmental stages:

• Experiment 1: A total of 188 juvenile fish from Vesterålen Havbruk were examined . No clearly developed gonads were observed in any of the juveniles. Hyperspectral images were taken from the ventral and lateral sides. Ultrasound is completely ineffective on such small individuals; therefore, sex could only be determined by dissection.

• Experiment 2: Analysis was performed on 136 larger fish from Kime Akva . Hyperspectral data were collected, but ultrasound showed a low accuracy of approximately 40% and was considered unreliable, requiring sex determination by dissection.

• Experiment 3: Thirty fish from the Nofima cod breeding program were included in a smaller, more detailed experiment. In addition to hyperspectral images, fluorescence images were captured, showing the light emitted by the irradiated tissues and highlighting differences between different tissue types. Gonads were also measured directly.

• Experiment 4: A total of 998 fish from the Nofima breeding program were photographed using standard RGB (standard color images based on red, green, and blue light), and hyperspectral images of the ventral side were also taken for a subset of the fish.

Researchers also tested whether fish surfaces exhibit sex-specific light emission (called biofluorescence), or whether computer models with wild-type artificial intelligence could discern differences in fish external features from ordinary images. These experiments also failed to yield any results.

A technological revolution with future potential

Despite current setbacks in juvenile cod, non-invasive spectroscopy has developed rapidly in recent years and is already transforming the food processing industry at a pace comparable to conveyor belts in industrial production.

This method is now actively used in commercial settings to determine melanin and blood spots in salmon, measure fat content in fish such as salmon, mackerel, and herring, detect defects in fish fillets, assess moisture content in dried cod and salted dried cod, evaluate the quality of chicken breast, and even measure the amount of crab meat in crab legs—a multi-million dollar industry.

The Nofima team remains optimistic about further expanding the boundaries of optical technology.

“Sex determination in the early and mid-growth stages is crucial for what we call selective breeding. Earlier and more accurate sex determination can lead to better broodstock management, more efficient breeding, and better production planning, thereby reducing resource use and contributing to improved sustainability in cod farming,” Ortega said. “Furthermore, although fluorescence imaging and conventional imaging did not yield reliable results in this study, these methods may still have greater potential for future improvements in image-based sex determination.”

editorial@seafood.media
www.seafood.media

Print