Intelligent fry collecting system based on ecological characteristics of tilapia and fry collecting method thereof

Abstract

The application discloses an intelligent fry collecting system based on ecological characteristics of tilapia and a fry collecting method thereof, and belongs to the technical field of aquaculture. The system comprises a fry density monitoring and spatial positioning subsystem, a multi-modal induction subsystem, a self-adaptive flexible catching subsystem, a fry nondestructive conveying and temporary storage subsystem and a central intelligent control subsystem. The fry density monitoring and spatial positioning subsystem generates a soil pool fry density grid map through underwater AI visual recognition. The multi-modal induction subsystem is composed of a light induction module, a bait induction module and a sound wave induction module, and cooperatively induces the fry to a catching point. The central intelligent control subsystem receives the density grid map and prestores a density threshold value and an induction duration. When the grid density exceeds the first threshold value, the induction is started, and after the induction reaches the preset duration, the catching is started. The application realizes efficient, low-damage and continuous soil pool fry collecting operation.

Description

Technical Field

[0001] This invention relates to the field of aquaculture technology, specifically to an intelligent fry catching system and method based on the ecological characteristics of tilapia. Background Technology

[0002] Nile tilapia is one of the most important freshwater farmed fish species in my country. During tilapia fry rearing, after the parent fish mate and spawn in earthen ponds, the fertilized eggs incubate in the female's mouth for about a week. When the newly hatched fry can swim horizontally and live independently, farmers need to collect them from the ponds using a seine net – an operation known in the industry as "fry harvesting." Currently, tilapia fry farms in my country generally use the traditional manual seine net harvesting method, which has the following significant drawbacks:

[0003] (i) Inefficient and incomplete harvesting. Traditional fry harvesting methods mainly rely on manual netting, which can only be done once a day under normal circumstances, making it impossible to remove all the fry from the earthen ponds. The fry left in the earthen ponds will prey on newly hatched fry, resulting in serious cannibalism and directly affecting the fry yield.

[0004] (ii) Severe damage to broodstock affects subsequent reproduction. During the broodstock and fry retrieval process, the broodstock and fry are caught in the net at the same time, resulting in injuries to the broodstock's body surface and a strong stress response, which seriously affects the broodstock's subsequent spawning performance and reproductive efficiency.

[0005] (iii) Fish fry suffer significant losses and are easily injured. The mechanical operation of manual netting causes the fish fry to squeeze and rub against each other in the net, resulting in the loss of mucus on the surface of the fish fry, damage to the scales, and even death, which reduces the quality and survival rate of the fry.

[0006] (iv) High labor intensity and high labor costs. The fry retrieval operation requires multiple people to work together to pull the net, which is time-consuming and labor-intensive, and must be carried out at specific times such as early morning or evening, which increases the labor cost burden of the farm.

[0007] (v) Lack of scientific design based on the ecological behavior of fish. Most existing fry collection methods are based on experience and do not make full use of the ecological behavior of tilapia, such as phototaxis, feeding tendency, and group swimming habits, to conduct scientific and precise design.

[0008] A search revealed the following limitations in existing patents related to tilapia fry harvesting: CN202020407065.9 (A Screenable Fry Capture Box) uses a static capture box, requiring manual placement and removal, and does not achieve automated continuous harvesting; CN201610656691.X (Tilapia Juvenile Collection Device and Method) is a passive collection method, relying on the active swimming of fry, and its collection efficiency is greatly affected by the activity level of the fry; CN201220466934.0 (A Trapping Device for Red Tilapia Juveniles) uses a double-layer funnel net combined with a trapping spotlight, but it does not combine bait trapping to form a dual trapping mechanism, nor does it involve AI visual recognition and automated control. None of the above-mentioned existing patents achieve intelligent and precise fry harvesting based on real-time monitoring of fry density. Summary of the Invention

[0009] To address the problems of low efficiency, damage to parent fish, high loss of fry, and high labor intensity in existing tilapia fry harvesting techniques, this invention provides an intelligent fry harvesting system and method based on the ecological behavior of tilapia, aiming to achieve efficient, low-damage, and continuous fry harvesting operations in earthen ponds.

[0010] An intelligent tilapia fry harvesting system based on the ecological characteristics of tilapia includes:

[0011] The fish fry density monitoring and spatial positioning subsystem includes multiple sets of underwater image acquisition units deployed in an earthen pond. Each set of image acquisition units acquires fish fry images of a corresponding grid area, and the image processing module performs target detection and multi-target tracking processing to identify and count individual fish fry. Based on the counting results of each grid area, a density grid map reflecting the fish fry density distribution in the earthen pond is generated. The fish fry density monitoring and spatial positioning subsystem is connected to a central intelligent control subsystem via a data signal line and transmits the density grid map to the central intelligent control subsystem.

[0012] The multimodal trapping subsystem includes a light-attracting module, a bait-attracting module, and an acoustic-inducing module, which are used to emit inducing light signals into the water, release bait, and emit low-frequency pulsed acoustic signals, respectively. The multimodal trapping subsystem is connected to the central intelligent control subsystem through a control signal line and receives start and stop control signals.

[0013] The adaptive flexible fishing subsystem includes a variable diameter net inlet and a negative pressure suction module. The variable diameter net inlet is equipped with a diameter adjustment mechanism to adjust the net area and mesh size. The negative pressure suction module is equipped with a flow control mechanism to adjust the suction flow rate. The adaptive flexible fishing subsystem is connected to the central intelligent control subsystem via a control signal line and receives parameter adjustment signals. The fish outlet of the adaptive flexible fishing subsystem is connected to the fish inlet of the buffer storage box of the temporary storage subsystem via a delivery pipeline for non-destructive fish fry delivery.

[0014] A fish fry non-destructive transport and temporary storage subsystem includes a transport pipeline and a buffer storage box. The transport pipeline connects the fish outlet of the negative pressure suction module to the fish inlet of the buffer storage box. The buffer storage box is equipped with a grading screen.

[0015] The system includes a central intelligent control subsystem, which is connected to each subsystem via signals and has a built-in data processing terminal. The data processing terminal receives a density grid map and pre-stores a first density threshold, a second density threshold, and a trapping duration. When any grid density value exceeds the first density threshold, the multimodal trapping subsystem is activated to perform trapping operations. After the trapping duration is reached, the adaptive flexible fishing subsystem is activated to perform negative pressure suction and seedling extraction operations. During the suction process, the suction flow rate is adjusted according to real-time density feedback. When the grid density value drops below the second density threshold, the seedling extraction and seedling extraction operations are stopped.

[0016] As a preferred embodiment of the present invention, the image acquisition unit is an AI vision camera group, which is arranged in groups at 5-meter intervals along the long axis of the earthen pool. Each group contains two cameras that monitor the surface water layer and the bottom water layer respectively. The AI ​​vision camera group includes an image sensor module and a supplementary lighting module. The supplementary lighting module includes a dual-color adjustable light source with a wavelength of 460nm blue light and 520nm green light.

[0017] In a preferred embodiment of the present invention, the image processing module uses the YOLOv8 model for target detection. The YOLOv8 model introduces a P2 small target detection layer to enhance small target detection capabilities, and adds a GAM attention mechanism before the detection head to improve detection accuracy. The target recognition loss function is the Inner-SIoU loss function to accelerate model convergence. The image processing module uses the DeepSORT multi-target tracking algorithm for multi-target tracking, which can solve the problem of maintaining the identity of fish fry when they are occluded or moving rapidly. The system generates a grid map of fish fry density in the earthen pond by fusing data from multiple cameras.

[0018] In a preferred embodiment of the present invention, the light-attracting module is an LED light-attracting lamp assembly comprising three wavelength light sources: a blue light source with a wavelength of 460-475nm and a peak value of 470nm, selected based on the fact that tilapia exhibit significantly higher phototaxis to blue light than to red, yellow, and green light; a near-ultraviolet light source with a wavelength of 365-400nm and a peak value of 385nm, utilizing the tilapia's attraction response to near-ultraviolet light; and a white light source with a color temperature of 5000K, adapting to the attraction requirements under different water quality and transparency conditions. The water surface illuminance is controlled within the range of 500-1500 lx.

[0019] The acoustic induction module is an underwater speaker. Taking advantage of tilapia's sensitivity to low-frequency sound waves, an underwater speaker is set up below the fishing point to emit intermittent low-frequency pulse sound waves with a frequency of 30-100 Hz, an intensity of 120-140 dB, and a duty cycle of 1:3, inducing fish fry to gather at the fishing point.

[0020] As a preferred embodiment of the present invention, the bait attracting module is an underwater pneumatic feeder, and the attracting bait it is filled with contains the following components by weight: 30-40 parts fish meal, 12-18 parts shrimp meal, 12-18 parts fermented soybean meal, 8-12 parts high gluten flour, 6-10 parts refined fish oil, 4-6 parts brewer's yeast powder, 2-4 parts spirulina powder, 1-3 parts compound attractant, 0.3-0.8 parts vitamin C phosphate, 1-2 parts sodium alginate, and 0.3-0.8 parts astaxanthin; the particle size of the attracting bait is 50-500 micrometers.

[0021] As a preferred embodiment of the present invention, the negative pressure suction module adopts a submersible axial flow pump or a vacuum fish suction pump; the flow control mechanism is a frequency converter, the frequency adjustment range of the frequency converter is 20-50 Hz, and the suction flow rate is 0.3-0.5 m / s.

[0022] As a preferred embodiment of the present invention, the fish fry non-destructive transport and temporary storage subsystem includes a transport pipeline connecting a negative pressure suction module and a buffer storage box, wherein the buffer storage box is equipped with a grading screen.

[0023] In a preferred embodiment of the present invention, the data processing terminal is an edge computing terminal, receiving updated data from the density grid map every 5 minutes. The first density threshold is 50 fish / cubic meter, the second density threshold is 10 fish / cubic meter, and the trapping time is 3-5 minutes. The central intelligent control subsystem also includes an IoT communication module, which uploads the fish catching data to the central management platform of the aquaculture farm via LoRa or 4G communication.

[0024] This invention also provides a tilapia fry harvesting method based on the ecological characteristics of tilapia, for use in the above-mentioned fry harvesting system, comprising the following steps:

[0025] Fish fry images are acquired by multiple underwater image acquisition units deployed in the earthen pond, and target detection and multi-target tracking are performed to identify and count individual fish fry. A density grid map reflecting the density distribution of fish fry in the earthen pond is generated based on the counting results.

[0026] Using a first density threshold as a standard, grids in the density grid map whose density values ​​exceed the first density threshold are identified as hotspots for fish fry aggregation.

[0027] For the hotspot area, the light-attracting module is activated to emit inductive light signals, the bait-attracting module is activated to release attracting bait, and the sound-wave-attracting module is activated to emit low-frequency pulse sound waves to attract the surrounding fish fry to the target fishing point.

[0028] After the induction period reaches the preset time, the variable diameter mesh opening is activated to adjust the mesh opening area and the negative pressure suction operation is started. During the suction process, the suction flow rate is dynamically adjusted according to the real-time density value of the hot spot area.

[0029] When the real-time density value of the hotspot area drops below the second density threshold, the seedling extraction operation shall be stopped.

[0030] The caught fish fry are transported through pipelines to buffer storage boxes for temporary storage in different grades.

[0031] As a preferred embodiment of the present invention, the bait collection module adopts an underwater pneumatic feeder to automatically release bait in the hot spot area. The amount of bait is calculated according to the real-time density value of the hot spot area, and the feeding rate is 0.5-2 grams per 10,000 fish per feeding. The particle size of the bait is 50-500 micrometers.

[0032] As a preferred embodiment of the present invention, the water temperature in the buffer storage tank is maintained at the same level as that in the earthen pond, the dissolved oxygen concentration is maintained at above 6 mg / L, and the temporarily stored fish fry are transferred every 4 hours.

[0033] By adopting the above technical solution, the present invention has the following beneficial effects:

[0034] 1. Significantly improved fry harvesting efficiency: Through AI vision real-time monitoring, multimodal attraction and closed-loop control of automatic harvesting, multiple fry harvestings can be achieved every day, with 6-8 harvestings per day. Each harvesting takes 10-15 minutes, increasing the fry harvesting rate from less than 50% in traditional methods to over 90%, and the fry yield is expected to increase by 50%-80%.

[0035] 2. Minimal damage to broodstock: The fry retrieval process involves precise suction only on grids of a specific density, without involving netting the entire pond. The broodstock remain undisturbed throughout the process, effectively ensuring their subsequent spawning performance and reproductive efficiency.

[0036] 3. Significantly reduced fish fry loss rate: The use of flexible netting combined with precise flow rate control and negative pressure suction avoids the damage caused by the squeezing and friction of fish fry in traditional netting. The damage rate of fish fry body surface is reduced from 15%-20% in traditional netting to less than 2%.

[0037] 4. Significantly reduced labor intensity: The seedling harvesting process is fully automated, and one operator can manage 3-5 earthen ponds at the same time. Each pond only requires 1 person × 1 hour / day, reducing labor costs by more than 70%.

[0038] 5. High degree of intelligence and strong adaptability: The system can automatically adjust its working parameters according to different fish fry sizes and different pond conditions, and has wide applicability. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the overall structure of the intelligent seedling retrieval system of the present invention;

[0040] Figure 2 This is a schematic diagram of the camera structure of the AI ​​vision camera group of the present invention;

[0041] Figure 3 This is a schematic diagram of the connection relationship of the multimodal trapping subsystem of the present invention;

[0042] Figure 4 This is a simplified schematic diagram of the fish fry density monitoring grid map of the present invention;

[0043] Figure 5 This is a control flowchart of the seedling retrieval method of the present invention;

[0044] Figure 6 This is a comparison chart of key production indicators between the intelligent seedling retrieval system of this invention and the traditional manual net-pulling method for seedling retrieval.

[0045] The markings in the diagram are explained as follows:

[0046] Figure 1 The components include: 1. Earthen pond; 2. AI vision camera group; 3. LED light-attracting lamp group; 4. Underwater pneumatic feeder; 5. Underwater speaker; 6. Variable diameter mesh port; 7. Negative pressure suction module; 8. Buffer storage box; 9. Edge computing terminal.

[0047] Figure 2 Components: 101. Main housing; 102. Top cover; 103. Image sensor module; 104. ISP processor; 105. Lens assembly; 106. Aluminum alloy lens bracket; 107. Sapphire protective window; 108. Fill light module; 109. Transparent lampshade; 110. Epoxy resin; 111. Mounting bracket.

[0048] Figure 4 In the diagram, the grid arrangement is simplified to 6 rows × 8 columns, totaling 48 grids. The scale is 1:400. Each grid in the diagram corresponds to an actual size of 2.5m × 1.5m. Different filling densities represent the fish fry density level (tails / m³) generated in real time by the AI ​​visual recognition system and updated every 5 minutes. The "285" marked in the diagram is an example of the measured density value of a certain grid.

[0049] Figure 6 In the study: experimental data came from 15 replicates in 3 ponds, P<0.001 (independent samples t test). Detailed Implementation

[0050] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0051] This invention is based on the following ecological and behavioral patterns of tilapia in its system design:

[0052] Regarding phototaxis, studies have shown that tilapia exhibit positive phototaxis within an illumination range of 250-2000 lx, with the phototaxis peaking at 1000 lx, and the optimal illumination range being 1000-1500 lx. The phototaxis varies significantly under different colors of light, with blue light showing a higher phototaxis than red, yellow, and green light. Based on this, this invention designs blue light with a wavelength of 460-475 nm as the primary attractant light source.

[0053] Regarding feeding behavior, Nile tilapia larvae feed on small zooplankton such as rotifers after hatching; they can start feeding on the fourth day after hatching, and can be fed formulated feed from 13 days of age. Juveniles and young fish show a clear feeding tendency towards live food and artificial micro-particle feed of specific particle sizes. The size of the feed particles should increase with the growth of the fish, but the diameter should not exceed two-thirds of the fish's mouth diameter at any stage.

[0054] Regarding group movement and swarming patterns, tilapia exhibit obvious group behavior, with fry often clustering in shallow areas of earthen ponds after initial horizontal swimming. In terms of individual development and diurnal activity rhythms, tilapia fry are frequently active during the day, and changes in light intensity and temperature affect their activity level within the water layer.

[0055] Example 1

[0056] like Figures 1 to 6 As shown, the present invention was tested in an outdoor earthen pond at Guangdong Weiye Tilapia Breeding Co., Ltd. for fry harvesting.

[0057] The experiment was conducted at the breeding base of Guangdong Weiye Tilapia Breeding Co., Ltd. in Maoming City, Guangdong Province. Pond 1 was 50m long, 30m wide, and 1.2m deep, with an area of ​​1500m². Water quality conditions were: temperature 28±1℃, pH 7.2-7.8, dissolved oxygen ≥5.0mg / L, and transparency 30-40cm. The experimental fish population consisted of 50 parent Nile tilapia, with a female-to-male ratio of 2:1. Newly hatched fry hatched naturally, with a size of 0.8-2.5cm.

[0058] System by Figure 1 Deployment:

[0059] AI vision camera group 2 is set up with 9 groups, each group contains two cameras to monitor the surface water layer and the bottom water layer respectively. They are arranged in 3 rows × 3 columns, and one group is arranged at a interval of 5 meters along the long axis of the earthen pool. The resolution is greater than or equal to 4K, the frame rate is greater than or equal to 60fps, and it has automatic supplemental lighting function.

[0060] The LED light-attracting lamp group 3 is set up in 5 groups. The LED light-attracting lamp group 3 is arranged in a grid pattern with a spacing of 3 meters in the earthen pool 1, containing blue light 460nm, near ultraviolet 385nm and white light 5000K.

[0061] Four underwater pneumatic feeders are configured to automatically release attractant bait in hotspot areas. The amount of bait released is calculated based on the real-time density value of the hotspot area, with a feeding rate of 0.5-2 grams per 10,000 fish per release. The particle size of the attractant bait is 50-500 micrometers.

[0062] Two underwater loudspeakers are set up to emit intermittent low-frequency pulse sound waves with a frequency of 30-100 Hz, an intensity of 120-140 dB, and a duty cycle of 1:3.

[0063] Three sets of variable diameter network ports are installed;

[0064] The negative pressure suction module 7 uses a submersible axial flow pump or a vacuum fish suction pump with a power of 1.5-3.7kW and a flow rate of 30-80m³ / h. The suction capacity of a single vacuum fish suction pump is 11t / h, the fish suction rate is about 270kg / min, and the tank vacuum gauge reading is 0 to -0.04MPa. The flow control mechanism of the negative pressure suction module 7 is a frequency converter with an adjustment range of 20-50Hz and a suction velocity of 0.3-0.5m / s. The delivery pipeline is made of PVC flexible hose with a diameter of DN100-DN150 and an inner wall smoothness Ra of less than or equal to 0.8μm to reduce friction damage to fish fry.

[0065] Two buffer storage tanks (8) are configured, each with a volume of 300L. The tanks are equipped with multi-layered damping screens with apertures of 2mm, 5mm, and 8mm to facilitate initial separation of the fish fry from the water. Each buffer storage tank (8) contains an aeration system and a circulating water system. The aeration system uses nanoporous aeration discs with an air supply of 0.5-1.0L / min·L to maintain the water temperature consistent with that of the earthen pond.

[0066] One central control cabinet contains an edge computing terminal 9. The edge computing terminal 9 has a built-in YOLOv8 model. The YOLOv8 model introduces a P2 small target detection layer and adds a GAM attention mechanism. At the same time, the target recognition loss function is set to the Inner-SIoU loss function. The algorithm used by the edge computing terminal 9 to perform multi-target tracking is the DeepSORT multi-target tracking algorithm.

[0067] The specific structure of AI vision camera group 2 is as follows: Figure 2As shown, it adopts an IP68 waterproof sealing design: the main body consists of a 316L stainless steel main shell 101, a 304 stainless steel upper cover 102, and a silicone rubber sealing ring forming a sealed cavity; the cavity integrates an image sensor module 103, an ISP processor 104, and a lens assembly 105. The image sensor module 103 uses a Sony IMX677, and the lens is fixed by an aluminum alloy lens bracket 106. A sapphire protective window 107 is set at the front end, which has scratch resistance and anti-bioadhesion performance; the outside of the main shell 101 integrates a 460nm / 520nm dual-color LED supplementary light module 108, with an illuminance of 100-3000lx and a supplementary light angle that can be adjusted within the range of 0-60°. A PC transparent lamp cover 109 is set on the outside, and the inside of the cavity is potted with epoxy resin 110 to achieve waterproof insulation; a waterproof cable connector is set on the top of the device to lead out the signal cable, and an L-shaped stainless steel mounting bracket 111 is set at the bottom. The camera has a detection range of 0.5-5.0m, an optimal working distance of 1.0-2.5m, and a minimum illumination of 0.01 lux for color and 0.001 lux for black and white, which can meet the needs of fish fry image acquisition under different water layers and transparency conditions in earthen ponds.

[0068] The variable-diameter mesh opening 6 comprises an adjustable truss structure and a mesh covering mounted on the adjustable truss structure. The adjustable truss structure is a scissor-type telescopic frame, composed of multiple sets of X-shaped cross links connected end-to-end with hinges. Its structural principle is similar to a scissor jack or an alligator-mouth mechanism, achieving overall telescopic movement by changing the cross angle between adjacent links. One end of the adjustable truss structure is fixed, and the other end is movable. The fixed end is hinged to a support at the bottom of the earthen pool, while the movable end can slide horizontally. The diameter adjustment mechanism is an electric push rod. The cylinder end of the electric push rod is connected to the fixed end, and the push rod end is connected to the movable end. When the electric push rod extends, it pushes the movable end away from the fixed end, increasing the angle between the X-shaped cross links, thus unfolding the adjustable truss structure and increasing the opening area of ​​the variable-diameter mesh opening 6. When the electric push rod retracts, it pulls the movable end closer to the fixed end, decreasing the angle between the X-shaped cross links, thus retracting the adjustable truss structure and reducing the opening area of ​​the variable-diameter mesh opening 6. By controlling the extension and retraction stroke of the electric push rod, the opening area of ​​the variable diameter mesh 6 can be dynamically adjusted within the range of 1-6 square meters. The netting material is made of ultra-high molecular weight polyethylene monofilament with a strength of 40g / den. The mesh size is adjusted to 0.5 mm or 1.0 mm according to the size of the fish fry. Newly hatched fry (0.8-1.5cm) use a 0.5mm mesh, while juveniles (1.5-3.0cm) use a 1.0mm mesh.

[0069] The connection architecture between the edge computing terminal 9 in the central control cabinet and the various subsystem devices is as follows: Figure 3As shown: The system is powered by DC24V / 48V. The control link connects the light-attracting module, bait-attracting module, acoustic-inducing module, image acquisition unit, and water quality sensor group via RS-485 bus and following the Modbus protocol. Data is transmitted uplink to the aquaculture farm management terminal via LoRa / 4G / Wi-Fi communication. Downlink control commands can be sent to each execution module. At the same time, each device sends back the operating status and abnormal alarm signals to the control terminal, forming a complete closed-loop control link.

[0070] The control logic of the central intelligent control subsystem is as follows: The AI ​​vision system collects real-time fish fry density data for each grid, updating it every 5 minutes. When the density of a grid exceeds the threshold of 50 fry / cubic meter, the central controller issues a trapping command. The light-attracting module near that grid is activated, providing 460 nm blue light at 1000 lx illuminance, and the feeding module is activated, with a feeding rate of 1 g / 10,000 fry. After 3-5 minutes of trapping, the harvesting module is activated, adjusting the net opening to a suitable area and initiating negative pressure suction at a flow rate of 0.3-0.5 m / s. Suction continues until the AI ​​vision monitoring shows that the density of that grid has dropped below 10 fry / cubic meter, at which point it stops. The system automatically records the fry harvesting data, including time, location, quantity, and efficiency, and enters the next monitoring cycle.

[0071] The bait attraction module employs an underwater pneumatic feeder 4. The bait loaded into the underwater pneumatic feeder 4 contains the following components by weight: 30-40 parts fish meal, 12-18 parts shrimp meal, 12-18 parts fermented soybean meal, 8-12 parts high-gluten flour, 6-10 parts refined fish oil, 4-6 parts brewer's yeast powder, 2-4 parts spirulina powder, 1-3 parts compound attractant, 0.3-0.8 parts vitamin C phosphate, 1-2 parts sodium alginate, and 0.3-0.8 parts astaxanthin. The fish meal is imported steam-cooked fish meal with a crude protein content greater than or equal to 65%; the shrimp meal is Antarctic krill meal with a crude protein content greater than or equal to 60%, containing dimethyl-β-propionate thiatin, which has strong attractant activity; the compound attractant is a mixture of glycine, alanine, and glutamic acid in a 3:1:1 ratio. The attractant bait has a particle size of 50-500 micrometers, where 50-200 micrometers corresponds to the mouth diameter of larvae and 200-500 micrometers corresponds to the mouth diameter of juveniles. The selection is based on the fact that tilapia's initial food consists mainly of small zooplankton such as rotifers. From 13 days of age, artificial formulated feed can be used. The diameter of the feed particles should not exceed two-thirds of the mouth diameter of the fish at each stage. An underwater pneumatic feeder 4 automatically releases feed in the light-attracting hotspot area. The feeding amount is calculated in real-time according to the fry density, with a feeding rate of 0.5-2 grams per 10,000 fry per feeding.

[0072] The seedling harvesting process is as follows:

[0073] Initialization phase: The system starts up, each AI vision camera group 2 completes self-test, the light group preheats in low light 200lx white light mode, the underwater pneumatic feeder 4 is filled with attractant bait, the buffer storage tank 8 is filled with pool water and oxygenation is turned on.

[0074] Monitoring and trapping stage: AI visual algorithms generate a grid map of fish fry density, identifying grid areas with a fry density greater than 50 fry / m³. For example... Figure 4 As shown, the identification results are 285 fish / m³ for grid (5,3) and 210 fish / m³ for grid (5,1). The system prioritizes processing the grid with the highest density, activates 460nm blue light attraction (1000 lx) and bait feeding (1.2 g / 10,000 fish, bait particle size 100-200 μm), and turns on the underwater acoustic wave guide (50 Hz, 130 dB) for 3 minutes.

[0075] During the fry extraction stage: After 3 minutes of attraction, the AI ​​visual display showed that the density of fish fry in the target mesh increased from 285 fry / m³ to 620 fry / m³. The harvesting module was then activated: the net opening was adjusted to 2.5m², the mesh size to 0.8mm, and the negative pressure suction pump was turned on (35Hz, flow rate 45m³ / h, flow velocity in the pipe 0.4m / s). Suction continued for approximately 8 minutes, automatically stopping when the AI-monitored mesh density dropped to 8 fry / m³.

[0076] Temporary storage and transfer stage: Fish fry are placed in buffer storage tank 8 and graded through 3 layers of sieves (2mm, 5mm, 8mm). Buffer storage tank 8 maintains a water temperature of 28±0.5℃ and dissolved oxygen greater than or equal to 6mg / L. Operators transfer the fry every 4 hours.

[0077] Comparison of experimental results and data

[0078] Between April and June 2026, a comparative experiment was conducted using the system of this invention and traditional manual netting for seedling retrieval in the same earthen pond. The results are shown in Table 2.

[0079] Table 2. Comparison test results of the system of the present invention and the traditional manual net-pulling seedling retrieval in the same earthen pond.

[0080]

[0081] The above data comes from a comparative experiment in three earthen ponds, with each method repeated 15 times. The data are presented as mean ± standard deviation (n=15). Independent samples t-tests showed that all indicators of the system of this invention are significantly superior to the traditional manual net-catching method (P<0.001), indicating that this invention represents a highly significant improvement in fry catching efficiency, fry protection, labor saving, and yield increase.

[0082] While the specific embodiments of the present invention have been described in detail above, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention, and modifications or variations without creative effort are still within the protection scope of the present invention.

Claims

1. An intelligent tilapia fry harvesting system and method based on the ecological characteristics of tilapia, characterized in that, include: The fish fry density monitoring and spatial positioning subsystem includes multiple sets of underwater image acquisition units deployed in an earthen pond. Each set of image acquisition units acquires fish fry images of a corresponding grid area, and the image processing module performs target detection and multi-target tracking processing to identify and count individual fish fry. Based on the counting results of each grid area, a density grid map reflecting the fish fry density distribution in the earthen pond is generated. The fish fry density monitoring and spatial positioning subsystem is connected to a central intelligent control subsystem via a data signal line and transmits the density grid map to the central intelligent control subsystem. The multimodal trapping subsystem includes a light-attracting module, a bait-attracting module, and an acoustic-inducing module, which are used to emit inducing light signals into the water, release bait, and emit low-frequency pulsed acoustic signals, respectively. The multimodal trapping subsystem is connected to the central intelligent control subsystem through a control signal line and receives start and stop control signals. The adaptive flexible fishing subsystem includes a variable diameter net inlet and a negative pressure suction module. The variable diameter net inlet is equipped with a diameter adjustment mechanism to adjust the net area and mesh size. The negative pressure suction module is equipped with a flow control mechanism to adjust the suction flow rate. The adaptive flexible fishing subsystem is connected to the central intelligent control subsystem via a control signal line and receives parameter adjustment signals. The fish outlet of the adaptive flexible fishing subsystem is connected to the fish inlet of the buffer storage box of the temporary storage subsystem via a delivery pipeline for non-destructive fish fry delivery. A fish fry non-destructive transport and temporary storage subsystem includes a transport pipeline and a buffer storage box. The transport pipeline connects the fish outlet of the negative pressure suction module to the fish inlet of the buffer storage box. The buffer storage box is equipped with a grading screen. The system includes a central intelligent control subsystem, which is connected to each subsystem via signals and has a built-in data processing terminal. The data processing terminal receives a density grid map and pre-stores a first density threshold, a second density threshold, and a trapping duration. When any grid density value exceeds the first density threshold, the multimodal trapping subsystem is activated to perform trapping operations. After the trapping duration is reached, the adaptive flexible fishing subsystem is activated to perform negative pressure suction and seedling extraction operations. During the suction process, the suction flow rate is adjusted according to real-time density feedback. When the grid density value drops below the second density threshold, the seedling extraction and seedling extraction operations are stopped.

2. The intelligent fry harvesting system based on the ecological characteristics of tilapia according to claim 1, characterized in that: The image acquisition unit is an AI vision camera group, arranged at 5-meter intervals along the long axis of the earthen pool. Each group contains two cameras that monitor the surface water layer and the bottom water layer respectively. The AI ​​vision camera group includes an image sensor module and a supplementary lighting module. The supplementary lighting module contains a dual-color adjustable light source with a wavelength of 460nm blue light and 520nm green light.

3. The intelligent fry harvesting system based on the ecological characteristics of tilapia according to claim 1, characterized in that: The image processing module uses the YOLOv8 model for target detection. The YOLOv8 model introduces a P2 small target detection layer and adds a GAM attention mechanism, while setting the target recognition loss function to the Inner-SIoU loss function. The image processing module uses the DeepSORT multi-target tracking algorithm for multi-target tracking.

4. The intelligent fry harvesting system based on the ecological characteristics of tilapia according to claim 1, characterized in that: The light-attracting module is an LED light-attracting lamp group, which includes a blue light source with a wavelength of 460-475nm, a near-ultraviolet light source with a wavelength of 365-400nm, and a white light source with a color temperature of 5000K; the water surface illuminance is controlled within the range of 500-1500lx, and the LED light-attracting lamp group is arranged in a grid pattern with a spacing of 3 meters in the earthen pool; the sound wave induction module is an underwater speaker that emits intermittent low-frequency pulse sound waves with a frequency of 30-100 Hz, an intensity of 120-140 dB, and a duty cycle of 1:

3.

5. The intelligent fry harvesting system based on the ecological characteristics of tilapia according to claim 1, characterized in that: The bait attracting module is an underwater pneumatic feeder, and the attracting bait it is filled with contains the following components by weight: 30-40 parts fish meal, 12-18 parts shrimp meal, 12-18 parts fermented soybean meal, 8-12 parts high gluten flour, 6-10 parts refined fish oil, 4-6 parts brewer's yeast powder, 2-4 parts spirulina powder, 1-3 parts compound attractant, 0.3-0.8 parts vitamin C phosphate, 1-2 parts sodium alginate, and 0.3-0.8 parts astaxanthin; the particle size of the attracting bait is 50-500 micrometers.

6. The intelligent fry harvesting system based on the ecological characteristics of tilapia according to claim 1, characterized in that: The negative pressure suction module is a submersible axial flow pump or a vacuum fish suction pump, the flow control mechanism is a frequency converter, the frequency adjustment range of the frequency converter is 20-50 Hz, and the suction flow rate is 0.3-0.5 m / s.

7. The intelligent fry harvesting system based on the ecological characteristics of tilapia according to claim 1, characterized in that: The data processing terminal is an edge computing terminal, which receives updated data from the density grid map every 5 minutes; the first density threshold is 50 fish / cubic meter, the second density threshold is 10 fish / cubic meter, and the trapping time is 3-5 minutes; the central intelligent control subsystem is also equipped with an Internet of Things communication module, which is used to upload the fish catching data to the central management platform.

8. A method for catching tilapia fry based on their ecological characteristics, used in the fry catching system described in any one of claims 1-7, characterized in that, Includes the following steps: Fish fry images are acquired by multiple underwater image acquisition units deployed in the earthen pond, and target detection and multi-target tracking are performed to identify and count individual fish fry. A density grid map reflecting the density distribution of fish fry in the earthen pond is generated based on the counting results. Using a first density threshold as a standard, grids in the density grid map whose density values ​​exceed the first density threshold are identified as hotspots for fish fry aggregation. For the hotspot area, the light-attracting module is activated to emit inductive light signals, the bait-attracting module is activated to release attracting bait, and the sound-wave-attracting module is activated to emit low-frequency pulse sound waves to attract the surrounding fish fry to the target fishing point. After the induction period reaches the preset time, the variable diameter mesh opening is activated to adjust the mesh opening area and the negative pressure suction operation is started. During the suction process, the suction flow rate is dynamically adjusted according to the real-time density value of the hot spot area. When the real-time density value of the hotspot area drops below the second density threshold, the seedling extraction operation shall be stopped. And the caught fish fry are transported through pipelines to buffer storage boxes for grading and temporary storage.

9. The seedling harvesting method according to claim 8, characterized in that: The bait attraction module uses an underwater pneumatic feeder to automatically release bait in hotspot areas. The amount of bait is calculated based on the real-time density value of the hotspot area, and the feeding rate is 0.5-2 grams per 10,000 fish per feeding. The particle size of the bait is 50-500 micrometers.

10. The seedling harvesting method according to claim 8, characterized in that: The water temperature in the buffer storage tank is maintained at the same level as that in the earthen pond, and the dissolved oxygen concentration is maintained at above 6 mg / L. The fish fry are transferred every 4 hours.

Images