• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    Biomass estimation system in aquaculture based on optical sensors and neural networks. A system for estimating biomass in aquaculture based on optical sensors and neural networks that comprises two optical barriers (100, 200) identical to each other, where each optical barrier (100, 200) comprises, in turn, a first emitter block (1) of photoemitters in the infrared spectrum and a second receptor block (2) of photoreceptors in the infrared spectrum and means of identification of the fish by radio frequency, in such a way that the univocal identification of each fish is produced when passing through the barriers optical (100, 200) thanks to the radiofrequency identifier, for a subsequent classification of the fish identified by neural networks. (Machine-translation by Google Translate, not legally binding)
    公开号:ES2786798A1
    申请号:ES201930332
    申请日:2019-04-11
    公开日:2020-10-13
    发明作者:García Daniel Pérez；Martín Francicsco Javier Ferrero；García Ignacio Alvarez；Llopis Marta Valledor；Rodríguez Juan Carlos Campo；Blanco Juan Menéndez；González José Ramón Blanco
    申请人:Universidad de Oviedo；
    IPC主号:
    专利说明:

    [0002] Biomass estimation system in aquaculture based on optical sensors and neural networks
    [0004] The present invention relates to a system for estimating the biomass necessary for feeding in the technical field of aquaculture. The estimation is carried out by using optical sensors and neural networks to establish the optimal amount of biomass for feeding fish in intensive aquaculture facilities with the smallest possible margin of error.
    [0006] State of the art
    [0008] The reduction of feed costs in aquaculture is essential to achieve the sustainability of the industry, with great potential, both in reducing costs per feed unit and through the adoption of adequate feed management strategies.
    [0010] In intensive aquaculture facilities, feeding is carried out by supplying several daily feed rations, where the optimum ration supposes a daily weight of feed of around 4% of the biomass present, depending on the species and other environmental conditions. Errors in estimating biomass imply that daily feeding is not optimized, which can lead to overfeeding or underfeeding of the crop and, therefore, in any case, an increase in crop costs.
    [0012] The biomass estimation systems that currently exist on the market have not been able to reach stable accuracies with sufficiently small margins of error. The reasons why biomass is not calculated correctly are complex and include a combination of the technology used for counting and the size of the fish, the misuse of counting technologies, and even biological and environmental reasons. A medium-term objective is to reach an uncertainty of 0.1% in the fish counting system and 1% in the estimation of biomass. A reduction within these margins has a great impact on the economic performance of aquaculture companies. Thus, for example, an error of 5% in the estimation of biomass supposes, only in the European industry of salmonids, sea bream and seabass, approximately 91 million dollars annually.
    [0014] To achieve the proposed objective, that is, to achieve error figures equal to or less than 0.1% in the fish count and 1% in the estimation of biomass, it is not only necessary to improve current control practices, as well as As a better use of existing instrumentation, it is also necessary to improve the measurement and control methods used today.
    [0016] An example of the improvements being made in technology is the method and system described in WO2019 / 002881 A1. This document relates to a method and apparatus for providing a dynamic decision-making process in relation to feeding animals in water. More particularly, the invention described in this document refers to a method and an apparatus whose object is to improve the feeding and / or cultivation strategies used in an aquaculture installation that uses a plurality of sensors and artificial intelligence techniques to optimize feeding the fish.
    [0018] There are two commercial solutions for biomass estimation: Vaki Biomass Daily (from Pentair) and Varó Aqua's Biomass Measurement Frame (from Storvik). Both products use a frame, which contains two infrared light curtains. When the fish passes through the frame, its height is calculated from the infrared beams that are cut, while its length is estimated from the time of passage and the speed of passage of the fish. A model usually applied in marine farms to characterize the relationship between fish length ( l) and mass (m) is m = ql3, where q is an empirical parameter characteristic of the species. An alternative model that includes the height (h) of the fish is m = chl2 where c is another empirical parameter characteristic of the species. These fish measurement methods undoubtedly contribute to the error or uncertainty in the biomass calculation.
    [0020] Infrared light frame technology is well established, although for fish weight estimation it has been shown that it has a very low accuracy and that for intensive aquaculture it is not a suitable technique, since the measurements are notably far from reality. Although the catalogs of this equipment indicate an accuracy of the method greater than 90%, these percentages are only achieved under ideal conditions, when the fish are moving in a straight line and at a moderate speed. However, when the fish moves too fast or too slow, the measurement becomes distorted and moves away from reality.
    [0021] Therefore, the technical problems that have been detected in the state of the art and that the present invention seeks to solve are the following:
    [0023] a) The speed of the fish may not be constant, the swimming pattern may be deviated by local currents and the fish may even stop within the frame. The ability to measure the speed of the fish is a technological aspect that needs to be improved in the frame.
    [0024] b) The distance between the emitters and receivers of infrared light is approximately 1 centimeter, which translates into a very low resolution, especially for small fish and, consequently, in a measurement error. This measurement error is propagated in the calculation of the weight of the fish, because in these equipments the height of the fish is measured and the weight is estimated based on a mathematical formula that depends on the length cubed, the height and a intrinsic factor of the fish species itself.
    [0025] c) The measurement of the height of the fish, which is the parameter from which the weight is estimated, can be conditioned if the movement of the fish is not perpendicular to the frame. If the fish enters or exits the frame at an angle, the height of the fish may be overestimated. These equipment have problems when, through the frame, several fish or other elements - false positives - pass through the frame that can distort the silhouette and induce an error in the measurement of the fish. All this requires evaluating a large number of fish to achieve a good estimate of weight. Finally, the Storvik system is not an online system, so it cannot be used for feed control in aquaculture.
    [0027] With the system object of the invention, the aim is to solve the technical problems indicated by using new optical systems, low-cost electronics and algorithms with neural networks.
    [0029] Explanation of the invention
    [0031] In aquaculture, as indicated, the cost of feeding represents a significant percentage of the operating costs in a fish production farm, representing approximately 45% of the total operating costs. In intensive aquaculture facilities, feeding is carried out by supplying several daily rations of feed, the optimal ration totaling a daily weight of 4% environment food of the biomass present.
    [0033] It is an object of the present invention to provide a system for estimating biomass in an aquaculture facility with stable precision and a sufficiently small margin of error. This object is achieved by means of the invention as defined in claim 1. Other aspects of the invention are described in other independent claims. Preferred or particular embodiments of the various aspects that make up the present invention are defined in the dependent claims.
    [0035] More specifically, the biomass estimation system developed consists of two infrared light curtains. These light curtains act as a kind of "scanner" so that, when the fish passes through it, the light emitted by the emitters does not reach the receivers and, in this way, it is possible to reconstruct the silhouette of the fish. To obtain an image of the fish with an adequate resolution, it is required that the emitter curtains be very close and that both the emitters and the receivers be as close as possible, in a practical embodiment of the order of 5 mm.
    [0037] The activation of the emitters and receivers is carried out sequentially, controlled by a microcontroller. When the fish passes through the first curtain, the height of the fish is obtained. To obtain the length of the fish it is necessary to know its speed, which can vary, and the time it takes to reach the second curtain. Hence, the closer the light curtains are to each other, the smaller the error made in measuring the speed of the fish.
    [0039] The weight of the fish is obtained from its dimensions, width and height, and a parameter that depends on the type of species. To concentrate the light beam on the receivers, new lenses have been designed that are placed in front of the emitters. This is important, since when the receivers receive more light, the attenuation of the light as it propagates in the water is compensated.
    [0041] The system of the invention must solve the technical problem related to the identification of fish when two or more fish or other objects pass. To discriminate these types of situations, the 2D image of the fish is digitally processed, which allows the fish to be identified in real time. To do this, a biomass estimation method is carried out based on the height and width of the fish.
    [0042] Therefore, the system of the invention proposes a higher resolution optical system. The optical system, although it is also based on a framework of optical emitters and receivers, provides significant advantages over existing products. Current technology makes it possible to use smaller and higher power light sources and receivers. With this, it is possible to increase the resolution - by being able to place the emitters and receivers at a shorter distance from each other - and obtain a more realistic fish profile. This also contributes to the fact that the emitters and receivers are covered with convex cylindrical lenses made of plastic material that concentrate the light on the receivers, reducing the proportion of scattered light.
    [0044] The estimation of the biomass present in the cage in real time is a technical advantage that derives from the fact that the barriers have an antenna embedded that, together with a reader, constitutes a radio frequency system (RFID) that allows to identify in real time the fish that is passing through the barriers at every instant of time. This novelty, not present in commercial equipment, provides a substantial improvement in the estimation of biomass, since it allows knowing in real time the biomass present in the cage and, based on it, managing the automatic feeding system online.
    [0046] Finally, the system of the invention is configured to classify fish conditions by means of neural networks. The objective is the univocal identification that the object that passes through the barriers is a fish, ruling out false positives. To do this, neural networks have been trained that, in collaboration with the RFID system, make it possible to discard objects that are not fish, that is, false positives. In current systems, this task is performed, at best, by an operator from video recordings obtained using a camera.
    [0048] Throughout the description and claims, the word "comprises" and its variants are not intended to exclude other technical characteristics, additives, components or steps. For those skilled in the art, other objects, advantages and characteristics of the invention will be derived in part from the invention and in part from the practice of the invention. The following examples and drawings are provided by way of illustration and are not intended to restrict the present invention. Furthermore, the invention covers all the possible combinations of particular and preferred embodiments indicated herein.
    [0050] Brief description of the drawings
    [0051] A series of drawings that help to better understand the invention and that expressly relate to an embodiment of said invention, which is illustrated as a non-limiting example thereof, will now be described very briefly.
    [0053] Figure 1 shows a block diagram of the biomass estimation system in aquaculture based on optical sensors and neural networks, object of the present invention.
    [0055] Figure 2 shows, schematically, a view of the optical barriers that make up the system for estimating biomass in aquaculture, according to a particular embodiment of the invention.
    [0057] Figure 3 shows a flow chart with the fish biomass calculation algorithm according to one aspect of the present invention.
    [0059] Figure 4 shows a graph with the results of the estimation of the biomass of the fish according to the calculation algorithm shown in figure 3 estimating the mass from the length.
    [0061] Figure 5 shows a graph with the results of the estimation of the biomass of the fish according to the calculation algorithm shown in Figure 3, estimating the mass from the silhouette area.
    [0063] Detailed description of a practical embodiment of the invention
    [0065] As can be seen in the attached figures, the system proposed by the present invention comprises two identical optical barriers. Figure 1 shows the block diagram of the system object of the present invention. Blocks 1 and 2 represent, respectively, the first emitter block 1 and the second receiver block 2 of light in the infrared spectrum. These emitter 1 and receiver 2 blocks, in a practical embodiment, are separated by a maximum of 300 mm.
    [0067] The first emitter block 1 comprises a plurality of photoemitters that -in a non-limiting practical embodiment- consist of 96 high-power micro LEDs (light-emitting diodes) whose emission in the infrared spectrum is centered at 940 nm. Activation and the individual control of each of these light emitters is carried out from a signal conditioning circuit 3 that comprises, at least, one digital multiplexer with serial interface and a plurality of driver circuits configured to increase the output current of the, at least one, digital multiplexer and thus be able to activate the light emitting diodes of the first emitter block 1.
    [0069] The second receiver block 2 comprises a plurality of photoreceptors that consist -in a practical non-limiting embodiment- of 96 phototransistors (at least one photoreceptor for each photoemitter) whose detection band in the infrared spectrum is between 870 nm and 950 nm. . The output of the phototransistors is directly connected to at least one analog-to-digital converter circuit 4 (CAD). This analog-digital converter circuit 4 in practical embodiments can be an independent circuit or be integrated in a microcontroller 5.
    [0071] The CAD circuit 4 has, in this practical embodiment, eight inputs and, therefore, they are connected and capable of reading the measurement of up to eight phototransistors. The resolution of the CAD circuits 4 is twelve bits. Thanks to the use of detectors with analog reading, a better resolution is achieved in the detection of objects, since it allows interpolating measurements on the edges of the objects detected, giving rise to an effective resolution that improves the use of the 96 phototransistors used in this embodiment. practice.
    [0073] Each light barrier 1 and 2 is controlled by a microcontroller 5 that is connected in series, in turn, with a central processing unit 6.
    [0075] The microcontroller 5, in a practical embodiment, comprises a processor core and at least one memory where a program or programs composed of instructions are stored which, when executed by the processor core, cause the microcontroller 5 to generate a timed sequence of power-on and reading of the light barriers 100 and 200, and the serial communication with a central processing unit 6. In a non-limiting practical example, this microcontroller 5 is an «Arduino Nano» that is responsible for generating the light-emitting and light-emitting sequence. reading the photoreceptors associated with each photoemitter, in a sequence with controlled timings.
    [0077] Thus, the values read in the CAD circuits 4 are transmitted, through a virtual serial port protocol at 115200 baud to the central processing unit 6. This unit 6, in a practical example, it is a Raspberry Pi class B SBC that has four integrated USB ports, which allows four serial communications to be established with as many microcontrollers 5, in short, with each central processing unit 6 it is possible to control as many light barriers 100 and 200 as serial ports are available, in this practical example, up to four.
    [0079] The central processing unit 6 is housed in an industrial enclosure with IP68 protection, although it could be another enclosure with an even higher degree of protection as long as it can be installed together with barriers 1 and 2, submerged in water.
    [0081] To maximize the data transfer rate of the communications between the microcontroller 5 and the central processing unit 6 a digital protocol is used. Analog readings are transmitted by a continuous stream of data using only 8 bits (1 byte) per receiver block 2. The character 0xFF (d255) has been reserved to indicate the ends of the data string and therefore the rest of the conversion results to a maximum value of 0xFE (d254). This protocol achieves a data transfer of less than 2 ms per receiver block 2, with 96 individual photoreceptors.
    [0083] Finally, in order to minimize the number of connections required, the system has been designed to communicate with the outside through Ethernet communications, the power being carried out through POE ( "Power over Ethernet"). An external injector provides the POE power on the Ethernet cable and in the enclosure where the central processing unit 6 is housed, a “splitter” has also been included that supplies it with the 5 Vdc it needs to operate from the POE power supply. In this way, only a single cable has to come out of the water, with communications and electrical power.
    [0085] Figure 2 shows the physical construction of the light barriers 100 and 200. The emitter 1 and receiver 2 blocks are divided into modules 7, where each module 7, in turn, is divided into three mutually identical printed circuit boards, each one of which has 32 photoemitters or photoreceptors, totaling the 96 photoemitters or photoreceptors of this practical example of implementation of the invention. The separation between each of the photoemitters or each of the photoreceptors is a maximum of 4 mm, while the separation between each emitter block 1 or each receiver block 2 is a maximum of 100 mm.
    [0086] The printed circuit boards are housed inside a watertight enclosure 8 made of high-density black PVC. The lid of the watertight enclosure 8 is a sheet of methacrylate 9 that acts as a filter for infrared light coming from the outside. From each module 7 a single cable 10 with ten conductors comes out, including the power terminals. Cable glands are used to seal the cable against ingress of water.
    [0088] The emitter 1 and receiver 2 blocks are each provided with a convex cylindrical lens 11 made of plastic material that is configured to concentrate the light on the receivers 2, reducing the proportion of scattered light. This improves the signal-to-noise ratio at the output of the CAD circuit 4, in addition to increasing the separation distance between modules 7.
    [0090] The enclosure of barriers 1 and 2 is embedded with an antenna 12 that is used for the identification of fish by radio frequency (RFID). This antenna operates in the 134.2 kHz low frequency band with a range of up to 500 mm. Each of the fish carry a passive tag (PIT tag) that contains a unique identifying number. The tag is activated when the fish passes near the antenna, sending the unique identifying code to the RFID reader that is located on the outside of the raft.
    [0092] Figure 3 shows the algorithm for calculating the biomass of the fish. The first task is to calibrate each receiver block 2, since the sensitivity can change a lot between each pair of photoemitter and photoreceptor. For this, the reading stages of the two light barriers 100 and 200 of figure 2 are implemented. The reading stages are referenced as 30.1 and 30.2, it being understood that all the references of figure 3 are indicated with ".1" or ". 2 "refer to the measurements taken in the receiver blocks 2 of the two light barriers 100 and 200 represented in figure 2.
    [0094] Calibration is performed by compensating the 2D image using the gain and offset values of each photoreceptor. To measure the «offset», a dark frame is used, obtained with the emitters off. The gain is measured by means of a flat shot obtained with the emitters activated and without obstacles (“flat frame”). The process of obtaining calibration shots 31.1, 31.2 should be repeated periodically to compensate for the effects produced by variations in ambient lighting conditions, dirt on the optical surfaces, and changes in water turbidity. To do this, the flat shot is dynamically updated by applying a median average to a selection of the last uncalibrated shots. The dark shot is updated from the same
    [0097] mode, that is, from the last unlit shots.
    [0099] Each calibrated tap 31.1 and 31.2 consists of a list of 96 values (in this particular non-limiting embodiment) of "shadow", between zero and one. The zero value indicates that there is no occlusion between emitter and receiver, while the value one corresponds to a sensor blocked with an interposed object. The last shots of each light curtain 1 and 2 are stored in FIFO stacks 32.1 and 32.2. The size of the stack is previously defined based on the frequency of reading the light barriers 100 and 200 and an estimate of the time it may take for a fish to transit in front of it.
    [0101] The shots stored in each FIFO queue 32.1, 32.2 can be interpreted as a two-dimensional image of the fish passing through the light barriers 100 and 200, where the vertical axis represents the position of each sensor (each photoemitter-photoreceptor pair) and the horizontal axis the instant in which the shot was obtained.
    [0103] The silhouettes that may exist in the FIFO queues are then found and tagged using a segmentation algorithm. The result is a collection of segments 33.1 and 33.2 that are small rectangular images that contain the silhouette of a fish. Segments 33.1 and 33.2 of each light barrier 1,2 are paired taking into account their position and their instant of detection, so that, to each transit of a fish there are two segments 33.1 and 33.2, one for each light barrier 1 and 2. From the two segments 33.1 and 33.2 and taking into account the detection instants of both, the speed of the fish is estimated, which in turn is used to obtain a single combined segment 34 to which the horizontal axis must be corrected. 36. Thus, each combined segment 34 is labeled 35 with the instant of time in which it has been detected and the unique identifier obtained by the radio frequency antenna 12. Combined segments 34 that cannot be uniquely identified with a unique RFID identifier 35 are discarded. Finally, each segment combined 34 and labeled 35 must be corrected for length. The correction of the horizontal axis 36 consists of a scaling of the horizontal dimension of the image, so that the horizontal scale (pixels per unit length) is the same as the vertical one.
    [0105] Subsequently, a neural network classifies 37 each corrected segment 36 into different categories that represent the different conditions that can affect the measurement. The neural network is pre-trained by classifying a set of segments by hand. The possible categories, in this example of practical realization are: «good», 'Slanted', 'mixed', 'distorted' and 'unrecognizable'. Segments not classified as "good" can be discarded or stored in an external database for later examination by an operator.
    [0107] Two different threads can be used to estimate the mass of fish 38 from the segments categorized as "good" by neural network 37:
    [0109] a) From the length. There are several alternative models to estimate the mass of the fish based on its length. A general model is of the form M = f ( L) = aL b, where M is mass, L is length, and a, b are empirical parameters that depend on the species. Figure 4 shows the results obtained by applying this model to rainbow trout.
    [0110] b) From the area of its silhouette. To do this, it is enough to add the value of each pixel in the segment and apply, as in the case of length, a formula that relates area to mass. A general model is of the form M = f ( Á) = cA d where M is the mass, A is the area of the silhouette and c, d are empirical parameters that depend on the species. Figure 5 shows the results obtained by applying this model to rainbow trout.
    [0112] Finally, the valid segments with their identifiers and their classification are entered in an external database 39, which communicates via Ethernet with the central processing unit 6, which is the device that executes the indicated method and in which they will also be periodically stored. records on water quality, quantity of feed supplied and other data relevant to the aquaculture facility.
    权利要求:
    Claims (15)
    [1]
    one.
    [2]
    two.
    [3]
    3. method according to any one of claims 1 or 2 comprising a stage of communication via Ethernet with an external database (39) where at least the valid segments with their identifiers and their classification are stored.
    [4]
    Four.
    [5]
    5.
    [6]
    6.
    [7]
    7.
    [8]
    8. - The system according to any one of the preceding claims, wherein the photoreceptors of the second receptor block (2) are phototransistors whose detection band is between 870 nm and 950 nm.
    [9]
    9.
    [10]
    10.
    [11]
    eleven.
    [12]
    12.
    [13]
    13.
    [14]
    14.
    [15]
    fifteen. system according to claim 14 where the separation between optical barriers (1,2) is a maximum of 300 mm, while the separation between consecutive emitting blocks (1) or receivers (2) is a maximum of 100 mm and the separation between photoemitters and photoreceptors within each module (7) is a maximum of 4 mm.
    one
    类似技术:
    公开号 | 公开日 | 专利标题
    US20200267947A1|2020-08-27|Arrangement and method for measuring the biological mass of fish, and use of the arrangement
    JP2012519277A|2012-08-23|Physical parameter estimation using 3D display
    ES2881617T3|2021-11-30|Dirt measuring device for photovoltaic arrays employing microscopic imaging
    US20080130005A1|2008-06-05|Optoelectronic apparatus and a method for its operation
    ES2786798B2|2022-02-08|Biomass estimation system in aquaculture based on optical sensors and neural networks
    US20130051696A1|2013-02-28|Systems and methods for imaging of falling objects
    WO2009008733A1|2009-01-15|Means and method for average weight determination and appetite feeding
    KR101942874B1|2019-01-28|Aquatic animals counter
    CN111526716A|2020-08-11|Method and system for fish ectoparasite monitoring in aquaculture
    US9536304B2|2017-01-03|Determining pathogens based on an image of somatic cells in a fluid sample
    ES2811263T3|2021-03-11|Optical apparatus and method for counting living organisms in a stream
    CN111491508A|2020-08-04|System for fish ectoparasite monitoring in aquaculture
    Lin et al.2018|Three-dimensional location of target fish by monocular infrared imaging sensor based on a L–z correlation model
    CN111511201A|2020-08-07|System for fish ectoparasite monitoring in aquaculture
    CN207395666U|2018-05-22|Fry length measuring instrument structure based on line array CCD
    ES2645504T3|2017-12-05|Determination of at least one characteristic of a vehicle
    CN207610704U|2018-07-13|A kind of photopic vision dynamic positioning system based on light stream
    CN111511203A|2020-08-07|Method and system for fish ectoparasite monitoring in aquaculture
    Zhao et al.2021|Estimating animal population size with very high‐resolution satellite imagery
    Ferrero et al.2014|Optical systems for the detection and recognition of fish in rivers
    ES2618735T3|2017-06-22|Device for measuring a body object of a slaughterhouse animal
    CN108303118B|2022-03-18|Space camera on-orbit radiometric calibration system and method based on active radiative illumination source
    US20220004760A1|2022-01-06|Splash detection for surface splash scoring
    US20200162656A1|2020-05-21|Method and apparatus for nonuniformity correction of ir focal planes
    Ohmura et al.2010|A real-time stereo vision sensor based on FPGA realization of orientation code matching
    同族专利:
    公开号 | 公开日
    ES2786798B2|2022-02-08|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    NO330863B1|2007-07-09|2011-08-01|Feed Control Norway As|Apparatus and method for cutting weight milling and appetite lining in fish farms|
    NO330423B1|2009-06-26|2011-04-11|Storvik Aqua As|Device and method for fish counting or biomass determination|
    NO20121541A1|2012-12-20|2014-06-23|Ebtech As|System and method for calculating physical sizes for freely moving objects underwater|
    CN106355589A|2016-09-20|2017-01-25|北京农业信息技术研究中心|Estimation system and method of factory-like circulating water cultured fish space|
    WO2018111124A2|2016-12-15|2018-06-21|University Of The Philippines|Estimating fish size, population density, species distribution and biomass|
    KR20180076083A|2016-12-27|2018-07-05|부산대학교 산학협력단|Aquatic animals counter|CN113197145A|2021-05-08|2021-08-03|浙江大学|Fish biomass estimation system based on cyclic neural network and infrared measurement grating|
    法律状态:
    2020-10-13| BA2A| Patent application published|Ref document number: 2786798 Country of ref document: ES Kind code of ref document: A1 Effective date: 20201013 |
    2022-02-08| FG2A| Definitive protection|Ref document number: 2786798 Country of ref document: ES Kind code of ref document: B2 Effective date: 20220208 |
    优先权:
    申请号 | 申请日 | 专利标题
    ES201930332A|ES2786798B2|2019-04-11|2019-04-11|Biomass estimation system in aquaculture based on optical sensors and neural networks|ES201930332A| ES2786798B2|2019-04-11|2019-04-11|Biomass estimation system in aquaculture based on optical sensors and neural networks|
    [返回顶部]