Classification of guppies’ (Poecilia reticulata) gender by computer vision

Labor costs of guppy growers and breeders are largely those of manual sorting (by strain, quality and gender) and counting fish. In most farms, female and male fish are grown together and sold either separately or together. Sorting fish according to gender is important for marketing as well as for breeding programs, so that a device for sorting and counting fish can potentially reduce production costs and improve quality.

A project aiming to develop sorting and counting technologies for ornamental fish growers included development and testing of image-processing algorithms for sorting guppy fish (Poecilia reticulata) by gender. The algorithms are derived from shape and color differences between female and male guppies. An algorithm for the determination of landmarks on fish contours was developed and found to be accurate in accordance with human judgment, enabling extraction of specific shape and color features of the tail and the body.

The algorithms were applied to three sets of images of guppies of the “Red-Blond” strain. Gender identification accuracy was approximately 90% using shape features, approximately 96% using color features and was slightly improved when both color and shape features were used.

Some of the components used are essential for future development of a computer vision based system for sorting and grading ornamental fish by strain and quality.     

Automated measurement of species and length of fish by computer vision

Trials of a computer vision machine (The CatchMeter) for identifying and measuring different species of fish are described. The fish are transported along a conveyor underneath a digital camera. Image processing algorithms: determine the orientation of the fish utilising a moment-invariant method, identify whether the fish is a flatfish or roundfish with 100% accuracy, measure the length with a standard deviation of 1.2 mm and species with up to 99.8% sorting reliability for seven species of fish. The potential application of the system onboard both research and commercial ships is described. The machine can theoretically process up to 30,000 fish/h using a single conveyor based system.     

A computer vision system to analyse the swimming behaviour of farmed fish in commercial aquaculture facilities: A case study using cage-held Atlantic salmon

Knowledge of fish behaviour plays an important role in aquaculture farm management. Video systems are the most common and cost-effective way of observing behaviours in commercial aquaculture operations. However long term observation is not feasible due to a limited ability to analyse footage manually. This paper describes preliminary findings obtained via computer vision software that was developed to automatically analyse fish movement and behaviours in aquaculture sea cages. Results show that the system is capable of detecting fish shapes in video recordings and from these shapes quantifying changes in swimming speed and direction continuously throughout the day. Also variations between days were detected and these may have been associated with the daily shift in the tidal cycle. The system has the potential to act as an alarm to farm operators, informing them about unusual fish behaviours on a continuous, real-time basis. It also has potential to assist in the evaluation of fish welfare.     

Monitoring the behavior of hypoxia-stressed Carassius auratus using computer vision

Behavioral variations of a school of fish (Carrasius auratus) in a tank were monitored using computerized vision while the fish were subjected to hypoxia stress conditions. Two video cameras viewed the tank, one from the side through a transparent wall and one from above. The images were processed by a PC-based image processing algorithm which provided three coordinates of the school's center of gravity, with the standard deviations of each coordinate, and the projected spatial distribution of the fish. Stressed fish differed from controls with the following behavioral responses: (i) the center of gravity moved upwards and horizontally away from the transparent wall; (ii) swimming speed was apparently reduced; (iii) there were strong fluctuations in all three directions associated with spreading and contracting of the school; (iv) the periodic amplitude of the motion in the vertical direction increased. This study might evolve into an automatic stress warning technique for fish farms, based on alterations in the behavior of fish when stressed.     

A study of metabolic rates of Poecilia reticulata Peters under different conditions

Abstract. The aim of this work was to investigate the factors that affect the metabolic rates of guppies, Poecilia reticulata Peters, by measuring the oxygen uptakes of guppies individually or in groups in closed vessels to simulate the actual packaging condition. Metabolic rates of guppies increased with the increase of temperatures. Anaesthetic, 2-phenoxyethanoll suppressed the oxygen consumption rates. Grouped fish also showed lower metabolic rates than individual fish. Light and starvation did not produce any effect. The pH of the water, ammonium and carbon dioxide concentrations had significant effects on the metabolic rates of guppies.     

Automated sorting for size,sex and skeletal anomalies of cultured seabass using external shape analysis

In aquaculture, automation of fish processing by computer vision could reduce operating costs, improving product quality and profit. Currently fish are mechanically sorted by size, but market constraints require that externally malformed fish be removed as well. Additionally fish farmers screen for sex, in order to exploit the higher growth potential of females. The aim of this study was the development of methodological tools applicable to the on-line sorting of farmed seabass (Dicentrarchus labrax, L.) for size, sex and presence of abnormalities. These tools are based on image analysis and utilizing outline morphometry (Elliptic Fourier analysis) combined with multivariate techniques (based on partial least squares modelling). Moreover, the integration of these techniques produce size estimation (in weight) with a better regression efficiency (r = 0.9772) than the commonly used log of the measured body length (r = 0.9443). The two partial least squares discriminant analysis models used to select sex and malformed fish also returned high discrimination efficiencies (82.05% and 88.21%, respectively). The implementation of a similar approach within an on-line sorting machine would allow for real-time live fish processing.     

Evaluation of codends with sorting grids, exit windows, and diamond meshes: Size selection and fish behaviour

This paper presents the results of an experiment aimed at improving the size selection of cod (Gadus morhua) and haddock (Melanogrammus aeglefinus) in the Nordic bottom trawl fishery. Three systems simulating commercial conditions were tested: a 135 mm diamond-mesh codend fitted with a 55 mm sorting grid (Sort-V); a 135 mm diamond-mesh codend fitted with two lateral exit windows; and a codend built entirely of 155 mm diamond mesh. The selectivity curves showed similar selection ranges (SR) for the three systems. For cod, the mean SRs were around 8 cm, while for haddock, they were around 6 cm. All the estimated l₅₀ values were far above the minimum landing sizes (MLS). For cod, the mean l₅₀s were 56.1, 53.9, and 60.7 cm for the sorting grid, exit windows, and diamond-mesh configurations, respectively, while for haddock, they were 50.2, 50.6, and 49.9 cm, respectively. Underwater observations revealed that most of the fish escaped from the grid and exit-window codends as soon as they reached the vicinity of the sorting device. In contrast, fish remained inside the diamond-mesh codend for a longer time, and consequently were more exposed to physical damage before escape. In addition, many fish escaped from this codend during haul back and these fish were not likely to survive due to the rapid changes in pressure. Thus, the selectivity results obtained with this type of codend should be interpreted carefully.     

An automatic fishing machine based on acoustic conditioning

An automatic fishing machine based on acoustic conditioning was developed and tested in a reservoir. The floating fishing machine is remotely and automatically controlled to operate an underwater speaker, a feeder and underwater cameras that monitor fish behavior in real time. An open net pen installed under the system can be automatically closed to capture gathering fish. The dynamics of fish populations in the open net pen was monitored and discussed. Fish were conditioned to associate an acoustic signal with food by classical conditioning. Periodic retraining was used to reinforce the conditioned behavior of fish freely swimming in the reservoir. Fish were captured by calling them to the fishing machine using the acoustic signal and automatically closing the net pen around them: 2.5 times more fish were captured in such capture events than in control events, in which the net pen was automatically closed without calling the fish.     

A review of flatfish behavior relative to trawls

Trawls harness the innate avoidance behavior of fish to affect their capture. As such, effective bycatch reduction relies, in part, upon knowledge of behavioral differences between target and non-target species. The behavior of flatfish during herding, net entry and passage through trawls differs substantially from that of many roundfish. These differences result from the unique body morphology of flatfish and the constraints this morphology places upon their natural predator avoidance and evasion tactics. Paramount, in this regard, is the intimate association between flatfish and the seafloor. Flatfish utilize a detection minimization strategy that combines burial, highly evolved cryptic capabilities and low activity. Additionally, the maximum sustained swimming speeds and endurance of flatfishes are low compared to most roundfish. As a result, flatfish typically respond to trawl ground-gear at shorter distances, remain closer to the seafloor during herding and herd in the net mouth for a shorter time, prior to net entry, than roundfish. As is the case for roundfish species, light and temperature influence flatfish reactivity and herding behavior. At low ambient light levels herding behavior in flatfish breaks down, as fish initially respond to ground-gear by rising off the bottom, which removes them from the ground-gear's zone of influence. Similarly, at low temperatures some flatfish display reduced herding. Differences in behavior between flatfish and roundfish have spurred development of selective flatfish trawls, with low rise head ropes and/or open or large mesh intermediates that allow rising roundfish to exit the net opening, while flatfish remain close to the net floor, passing into the net and back to the codend. Lastly, a conceptual design for a counter-herding device is presented, which takes advantage of the longer reactive distances of roundfish, and their tendency to herd farther of the seafloor, thereby allowing them to be herded out of the path of the net, while not influencing the normal inward herding of flatfish.     

Automatic mass estimation of Jade perch Scortum barcoo by computer vision

The aim of this study was to test and evaluate a 2D computer vision technique that estimates the mass of Jade perch Scortum barcoo swimming freely in a tank of a recirculation aquaculture system. The first step of this study, which is described in this paper, was to build up a relationship between the fish shape and its mass in order to be able to estimate the mass of the fish by vision techniques. A set of 120 images of fish outside the water was captured and different features were extracted by using computer vision techniques. Regression analysis was used on the training dataset in order to generate the best model that estimated accurately the mass of the fish. Single-factor regression equation using the area of the fish without considering the fin tail proved adequate for measuring the mass of Jade perch S. barcoo and revealed a coefficient of determination (R²) of 0.99. When applied to the evaluation dataset, the mean relative error was 6 ± 3% compared to the value measured by a weighing scale. This suggests that the calculated model can be used in a second step to estimate the biomass of fish moving freely in a tank without causing any stress or damage to the fish.     

Environmental and physiological conditions affecting Tetrahymena sp. infection in guppies, Poecilia reticulata Peters

Parasitic infections caused by Tetrahymena sp. constitute a serious problem in guppies, Poecilia reticulata. Tetrahymena was isolated from skin lesions of naturally infected guppies in a commercial aquaculture farm, cultured in vitro and used in subsequent experimental infections. In addition to guppies, angelfish, Pterophyllum scalare, platyfish, Xiphophorus maculates, and neontetra, Paracheirodon innesi, were susceptible, whereas tilapia (Oreochromis niloticus xO. aureus) was resistant. The ciliate had a high affinity for dead fish. Skin abrasion did not affect the infection, but fish with gas bubble disease exhibited a significantly higher infection than non-affected fish. Infection was significantly higher when fish were exposed to high levels of ammonia, high organic load and low water temperatures. Under shipment conditions, infection was significantly elevated. Full recovery was achieved at a low fish density. Results suggest that poor environmental and physiological conditions enhance infection with Tetrahymena sp.     

基于卷积神经网络的养殖鱼类品种识别模型

随着人工智能、大数据、机器学习、计算机视觉等技术的发展,卷积神经网络（CNN）越来越多地应用于图像识别领域,图像数据集的丰富性以及多样性对CNN模型的性能和表达能力至关重要,但现有的鱼类图像公共数据集资源较匮乏,严重缺少训练集以及测试集样本,难以满足深度CNN模型优化及性能提升的需要。实验以大黄鱼、鲤、鲢、秋刀鱼和鳙为对象,采用网络爬虫以及实验室人工拍照采集相结合的方式,构建了供鱼种分类的基础图片数据集,针对网络爬虫手段获取到的鱼类图像存在尺度不一、格式不定等问题,采用图像批处理的方式对所有获取到的图像进行了统一的数据预处理,并通过内容变换以及尺度变换对基础数据集做了数据增强处理,完成了7 993个样本的图像采集与归纳;在权值共享和局部连接的基础上,构建了一个用于鱼类识别的CNN模型,采用ReLU函数作为激活函数,通过dropout和正则化等方法避免过度拟合。结果显示,所构建的CNN鱼种识别模型具有良好的识别精度和泛化能力。随着迭代次数的增加,CNN模型的性能也逐步提高,迭代1 000次达到最佳,模型的准确率为96.56％。该模型采用监督学习的机器学习方式,基于CNN模型,实现了5种常见鱼类的鱼种分类,具有较高的识别精度和良好的稳定性,并且模型本身具有一定程度的鱼种推广可能性,为养殖鱼类的品种识别提供了一种新的理论计算模型。     

海河流域大清河水系的鱼类多样性

为了解大清河水系的鱼类种群多样性组成特点,于2018—2019年间对大清河水系进行多次鱼类调查,结合历史文献资料,梳理了大清河水系的鱼类种类共85种,其中自然分布的淡水鱼类共计8目17科59属78种。这些鱼类多为常见种,珍稀濒危物种占比低,可以充分代表海河流域鱼类区系,是大清河鱼类群落组成的重要特点。然而现状调查只能调查到42种,其中尚自然分布于流域的种类则降低到33种,且以小型鱼类、小个体鱼类为主。Margalef丰富度指数、Shannon-Wiener多样性指数、Pielou均匀性指数分别为1.98、1.14、0.81,同海河流域其他水系相比,表明大清河水系鱼类物种多样性一般 较低,且各区域分布较均匀。多角度考察显示,目前流域鱼类区系的完整性已被高度破坏,鱼类多样性呈现出显著的下降趋势。经济发展驱动的用水需求和大清河流域水资源量不足是导致鱼类多样性减少的主要原因,生境变化、水域污染、水工建设、过度捕捞等也是不可忽视的因素。因此保证一定生态流量和一定程度恢复河流连通性、加强水污染治理、开展人工增殖放流、加大禁捕力度等,都可改善流域鱼类多样性现状。     

Use of avoidance response by rainbow trout to carbon dioxide for fish self-transfer between tanks

Convenient, economical, and reduced labor fish harvest and transfer systems are required to realize operating cost savings that can be achieved with the use of much larger and deeper circular culture tanks. To achieve these goals, we developed a new technology for transferring fish based on their avoidance behavior to elevated concentrations of dissolved carbon dioxide (CO₂). We observed this behavioral response during controlled, replicated experiments that showed dissolved CO₂ concentrations of 60–120 mg/L induced rainbow trout (Oncorhynchus mykiss) to swim out of their 11 m³ “growout” tank, through a transfer pipe carrying a flow with ≤23 mg/L dissolved CO₂, into a second 11 m³ “harvest” tank. The research was conducted using separate groups of rainbow trout held at commercially relevant densities (40–60 kg/m³). The average weight of fish ranged from 0.15 to 1.3 kg during the various trials. In all trials that used a constant flow of low CO₂ water (≤23 mg/L) entering the growout tank from the harvest tank, approximately 80–90% of the fish swam from the growout tank, through the transfer pipe, and into the harvest tank after the CO₂ concentration in the growout tank had exceeded 60 mg/L. The fish that remained in the growout tank stayed within the area of relatively low CO₂ water at the entrance of the transfer pipe. However, the rate of fish transfer from the growout tank to the harvest tank was more than doubled when the diameter of the transfer pipe was increased from 203 to 406 mm. To consistently achieve fish transfer efficiencies of 99%, water flow rate through the fish transfer pipe had to be reduced to 10–20% of the original flow just before the conclusion of each trial. Reducing the flow of relatively low CO₂ water near the end of each fish transfer event, restricted the zone of relatively low CO₂ water about the entrance of the fish transfer pipe, and provided the stimulus for all but a few remaining fish to swim out of the growout tank. Results indicate that the CO₂ avoidance technique can provide a convenient, efficient, more economical, and reduced labor approach for fish transfer, especially in applications using large and well mixed circular culture tanks.     

First report on the isolation of immunoglobulin M of guppy, Poecilia reticulata, for the production of polyclonal antibodies

As guppies are one of the commercially important fish in freshwater ornamental aquaculture industry, it is important to gain an understanding of guppy immune response for infectious disease control. Till now, the number of study that examined the immune response of guppies is limited and effective tools for monitoring guppy antibody have not been reported. In this study, we successfully isolated guppy IgM using mannan-binding protein (MBP) affinity chromatography and produced specific polyclonal antibodies against guppy IgM heavy and light chains, that showed a molecular weight of approximately 74 and 23 kDa respectively. The produced polyclonal antibodies were used to develop an enzyme-linked immunosorbent assay (ELISA) and have demonstrated to be an effective tool for the detection and quantification of antigen-specific antibody of guppies immunized with Pseudomonas fluorescens. In conclusion, the produced anti-guppy IgM polyclonal antibodies should prove its future implications for immunology and epidemiology studies in guppies.     

鱼类细胞多孔性与肌肉品质关系的研究进展

鱼类肉品品质与肌肉组织学特性有显著的相关性,相关因素包括肌纤维直径、数量、密度等(即肌肉细胞多孔性参数)。本文综述了鱼类肉的品质与肌纤维数量、密度、胶原蛋白及肉pH等之间的一些研究状况。肌肉细胞多孔性受品种、营养状况、运动状况等多种因素影响,因此了解可能影响肌肉细胞多孔性的因素可以为实践中通过合理控制这些因素来改善鱼肉品质提供有益参考。     

浙江南部近海鱼类粒径谱特征

为了解浙江南部近海鱼类群落粒径结构的时空特征、季节变动及其影响因素,基于2019年4个季节的渔业资源调查数据,利用粒径谱、多维尺度排序等方法,研究了鱼类粒径谱特征及其时空变动.结果表明,全年Sheldon鱼类粒径谱形状大体为单峰型,鱼类粒级介于?2～11,最高值出现在6～7粒级,以小型鱼类为主.四季Sheldon鱼类粒...     

海上风电场对鱼类福利的影响研究进展

为减少电力行业碳排放,实现碳中和目标,风电作为重要的清洁能源之一,受到世界范围的高度重视。随着陆上风电不断开发利用,导致陆地风电场空间资源日益紧缺,人类开始将风电产业转向更为广阔的海洋。由于我国海岸线绵长且邻近电力负荷区域,近年来,海上风电场建设取得了飞速发展,海上风电将成为我国未来发展可再生能源主要方向之一。海上风电工程在建设和运用期间会产生噪音和电磁场污染,可能对海洋生物会带来一定的负面影响,对其开展研究和评估是确保风电场可持续发展的前提。本文总结了国内外海上风电工程对海洋鱼类影响的研究进展,结合我国实际需求,提出研究建议,为未来我国开展海上风电项目对生物影响的评估提供参考。     

工厂化循环水养殖系统中流场特性与鱼类互作影响的研究进展与展望

随着人口与经济的发展,水产养殖业在世界范围内迅速兴起,集约型工厂化循环水养殖因其高密度、低污染、高效率等独特的优势,契合水产养殖业绿色发展理念,已成为水产养殖转型升级的重要方向之一。水作为循环水养殖系统中重要的环境因子,其流态能够直接影响鱼类的生长及福利,同样,鱼类存在及运动也会影响到系统流态的构建。本文综合分析了循环水养殖系统中流场条件对不同鱼类生长发育及福利的影响,鱼类及其运动行为对养殖池内水动力条件及性能的影响,以及鱼类对养殖池内流场流态、水体混合等的影响。将研究鱼类运动对流场特性的影响方法主要归纳为实测法和数值研究,通过对比分析2种方法的优点和不足之处,并结合当前循环水养殖产业系统构建中的问题提出针对性方法建议,旨在为系统中水动力条件的设计拓展思路,促进循环水养殖产业流态构建向“鱼”与“水”兼顾的方向发展。     

Automatic Bluefin Tuna (Thunnus thynnus) biomass estimation during transfers using acoustic and computer vision techniques

In this work, acoustic and computer vision techniques are combined to develop an automatic procedure for biomass estimation of tuna during transfers. A side scan sonar working at 200 kHz and a stereo camera, positioned facing towards the surface to record the ventral aspect of fish, are set as acquisition equipment. Moreover, a floating structure has been devised to place the sensors between cages in transfers, creating a transfer canal that allows data acquisition while fish swim from donor to receiving cage. Biomass assessment is computed by counting transferred tuna and sizing a representative sample of the stock. The number of transferred tuna is automatically deduced from acoustic echograms by means of image processing techniques, whereas tuna size is computed from the stereo videos using our automatic computer vision procedure based on a deformable model of the fish ventral silhouette. The results show that the system achieves automatic tuna counting with error below 10%, achieving around 1% error in the best configuration, and automatic tuna sizing of more than 20% of the stock, with highly accurate Snout Fork Length estimation when compared to true data from harvests. These results fulfil the requirements imposed by International Commission for the Conservation of Atlantic Tunas for compliant transfer operations.