Male and female crab seed sorting control system based on multi-view image recognition

Abstract

The invention relates to the field of automatic sorting of juvenile crabs, in particular to a juvenile crab male and female sorting control system based on multi-view image recognition. Comprising a material conveying module, an image acquisition module, an image recognition module, a control module and an air blowing module. The material conveying module conveys the juvenile crabs through a speed regulation conveying belt and controls the juvenile crabs to leave the belt at the tail end to form flying movement; the image acquisition module synchronously acquires an upper visual angle image and a lower visual angle image of the same juvenile crab in the juvenile crab flying stage; the image recognition module analyzes the collected image based on the morphological characteristics of the abdominal fins and the ventral fins and outputs a female crab or male crab distinguishing result; the control module receives the identification result, triggers a time sequence and outputs an air blowing instruction to the air blowing module; the air blowing module controls an air nozzle to jet compressed airflow through a high-speed electromagnetic valve, airflow is applied to target juvenile crabs to change the flight path of the juvenile crabs, and male and female shunting is achieved. The system can automatically complete male and female crab seed recognition and sorting, and the problems of low manual recognition efficiency, poor accuracy, high-intensity operation and the like are solved.

Description

Technical Field

[0001] This invention relates to the field of automatic sorting of crab seedlings, specifically a crab seedling sex sorting control system based on multi-view image recognition. Background Technology

[0002] In the Chinese mitten crab (Eriocheir sinensis) farming industry, separating male and female crab larvae for rearing is of great significance. Male and female crabs mature at different rates, with females maturing earlier and males later. During peak market periods such as around the Mid-Autumn Festival and National Day, female crabs often reach the market first. If polyculture is adopted, harvesting must be done in batches, and the harvesting process can easily cause damage such as crushing and abrasion to immature male crabs in the same pond. Furthermore, male and female crabs differ in their nutritional needs, feeding behavior, and growth rate at different growth stages. Separate rearing allows for targeted feeding and management, thereby improving farming efficiency and economic benefits. Especially in terms of commercial value, one-year-old female crabs typically command significantly higher prices than male crabs, and the price difference widens further during the juvenile stage. If the separation of males and females can be completed during the juvenile stage (e.g., at a scale of approximately 300-500 crab larvae per pound), with male crabs removed and the focus on raising female crabs, it will be more conducive to improving output per unit area and overall profitability.

[0003] Currently, the sexing of crab larvae still mainly relies on manual operation. Sorting personnel usually distinguish between males and females by visually observing the morphological characteristics of the "umbilicus" (ventral carapace / umbilicus) on the abdomen of the crab larvae. This method has objective and significant technical drawbacks: First, the sorting efficiency is limited and the labor intensity is high. Even with skilled personnel working continuously for long periods, the sorting volume per unit time is still insufficient to meet the needs of large-scale aquaculture for rapid processing of large quantities of crab seedlings. Second, the identification process relies heavily on subjective human judgment, which easily leads to visual fatigue and causes the misjudgment rate to increase with the extension of working time, resulting in poor sorting quality stability. Third, due to the influence of individual crab seedling size, subtle characteristics, and posture changes, manual identification has a clear range of applicability to different sizes. For smaller crab seedlings in larger quantities (e.g., seventy or even more than one hundred seedlings per pound), manual sorting efficiency is even lower. In practice, such small-sized crab seedlings are not sorted, greatly reducing their utilization value. Fourth, the slow pace of manual sorting prolongs the time that crab seedlings are out of water or in unsuitable environments. Combined with operations such as turning and picking, it is easy to cause stress and damage to the crab seedlings, thus affecting the survival rate. Fifth, large-scale reliance on manual labor will significantly increase labor and management costs, making it difficult to meet the industry's demand for standardized, continuous, and low-cost sorting.

[0004] Furthermore, while existing technologies exist for sorting crabs or larger individuals, these devices primarily employ mechanical sorting methods such as grasping and manipulating. They are typically designed for larger individuals in relatively limited numbers, and their sorting speed, flexibility, and adaptability are insufficient to address the challenges of sorting crab larvae, which involve "massive numbers, small individuals, and the need for stable feature identification." Simultaneously, crab larvae exhibit uncertain postures and random surface orientations during transport and sorting. Relying solely on a single perspective or requiring manual alignment further reduces the feasibility and efficiency of automated identification. Therefore, a dedicated sorting and control system is urgently needed that can achieve stable imaging and sex determination even under high-speed crab larvae passage conditions, and can complete the sex separation process with minimal damage. This system aims to meet the comprehensive requirements of high efficiency, high accuracy, and high survival rates in large-scale aquaculture. Summary of the Invention

[0005] The purpose of this invention is to provide a crab larvae sex separation control system based on multi-view image recognition, so as to solve the technical problems mentioned in the background art.

[0006] Based on the above ideas, the present invention provides the following technical solution:

[0007] A crab larvae sex separation control system based on multi-view image recognition includes:

[0008] Material handling module, image acquisition module, image recognition module, control module, and air blowing module;

[0009] The material conveying module is used to control the operation of the conveyor belt, so that the crab seedlings leave the conveyor belt at the end of the conveyor belt and form an airborne motion;

[0010] The image acquisition module is used to acquire top-view and bottom-view images of the same crab seedling during its aerial movement.

[0011] The image recognition module is used to output the female crab identification result or the male crab identification result based on the upper-view image or the lower-view image;

[0012] The control module is used to receive the sex determination result and output a blowing trigger command;

[0013] The blowing module is used to control the high-speed pneumatic solenoid valve to open the nozzle and spray air according to the blowing trigger command, so that the airflow acts on the target crab seedling to change its flight trajectory, thereby achieving male and female separation.

[0014] The material conveying module causes the crab larvae to leave the conveyor belt at the end, creating an aerial motion. During this aerial motion, top and bottom view images of the same crab larvae are acquired. The image recognition module outputs the sex determination result, and the control module triggers the air blowing module to blow air at a fixed point on the target crab larvae to change its flight trajectory, thereby achieving automatic separation of male and female crab larvae. Compared with the existing technology that relies on manual visual recognition, this solution can complete the closed-loop control of recognition and sorting under continuous conveying conditions, reducing reliance on manual labor and subjective misjudgment, improving sorting efficiency and consistency. Furthermore, since the recognition occurs during the aerial stage, it helps to reduce the impact of inconsistent crab larvae flipping on the recognition stability.

[0015] Preferably, it also includes a feeding module, which includes a feeding hopper, and the feeding rate is adjusted by adjusting the opening of the bottom plate of the feeding hopper;

[0016] The feeding module also includes a vibrating feeder, which is located below the feeding hopper to vibrate the crab seedlings, loosen and evenly disperse them, and allow them to enter the conveyor belt in a dispersed state.

[0017] By setting up a feeding module including a hopper and a baffle plate, and controlling the feeding rate by adjusting the opening of the baffle plate, and simultaneously installing a vibrating feeder below the hopper to vibrate, loosen, and evenly disperse the crab seedlings, the crab seedlings enter the conveyor belt in a dispersed state. This effectively avoids the obstruction and misidentification caused by the crab seedlings agglomerating and stacking, and provides stable and repeatable feeding conditions for subsequent aerial capture and air-blowing sorting, thereby improving the overall feasibility and sorting accuracy of the system.

[0018] Preferably, the material conveying module controls the speed-regulating motor to adjust the speed of the conveyor belt, so that the crab seedlings are conveyed in a single layer or near-single layer on the surface of the conveyor belt and stably leave the conveyor belt to enter the aerial capture area.

[0019] The material conveying module adjusts the conveyor belt speed by controlling the speed-regulating motor, so that the crab seedlings are conveyed in a single layer or near-single layer on the conveyor belt surface and stably detach from the conveyor belt into the air capture area. This can reduce the interference of multiple crab seedlings sticking together / blocking each other on the acquisition and discrimination of images from the upper and lower perspectives while ensuring the processing throughput. It also improves the stability and predictability of the detachment trajectory, making it easier for the subsequent control module to calculate the trigger time and the air blowing hit, thereby improving the consistency and reliability of sorting.

[0020] Preferably, the image acquisition module includes image acquisition devices arranged vertically, used to capture images of the same crab seedling moving in the air from multiple perspectives and output the upper perspective image and the lower perspective image.

[0021] The image acquisition module uses image acquisition devices arranged vertically to capture images of the same crab seedling in mid-air from multiple perspectives and output images from the upper and lower perspectives. This allows for the acquisition of key discrimination information through complementary different perspectives, even when the crab seedling's posture is randomly flipped and the upper and lower surfaces are inconsistent. This reduces the probability of recognition failure caused by occlusion, reflection, or posture deviation from a single perspective, and improves the robustness and applicability of sex identification.

[0022] Preferably, the image acquisition module further includes an upper supplementary light source and a lower supplementary light source, which are used to cooperate with the upper image acquisition device and the lower image acquisition device to capture images during aerial capture.

[0023] By setting up upper and lower supplementary light sources in the image acquisition module and cooperating with the upper and lower image acquisition devices to form images, stable lighting can be provided under short-exposure conditions for aerial capture, reducing the impact of shadows, backlighting, reflections and ambient light fluctuations on image quality, thereby improving the contrast and clarity of the acquired images, reducing the risk of misjudgment introduced by image noise, and further improving the stability and repeatability of the sex identification results.

[0024] Preferably, the image recognition module uses the shape of the pelvic fins on the abdomen of the crab larvae as a feature to distinguish between males and females, and fuses the recognition results from the upper and lower perspectives to output the final discrimination result.

[0025] The image recognition module uses the shape of the pelvic fins on the abdomen of crab larvae as a feature to distinguish between males and females. It fuses the recognition results from the upper and lower perspectives to output the final judgment result. This allows key morphological features with biological distinguishability to be used as the basis for judgment. By using complementary information from multiple perspectives to achieve cross-verification, it reduces misjudgments caused by missing features or partial occlusion from a single perspective, improves the credibility and consistency of the final judgment result, and thus improves the accuracy of male and female sorting.

[0026] Preferably, the image recognition module matches the recognition results with a pre-stored multi-view sample database to output a sex determination result.

[0027] Matching the recognition results with a pre-stored multi-view sample database to output the sex determination results is beneficial for the structured storage of typical features of sex differences from multiple angles and the formation of a callable discrimination basis. This enables fast and reproducible discrimination output under different batches, different specifications, or different shooting conditions, reduces reliance on operator experience, and facilitates subsequent expansion and updates of the sample database to continuously improve recognition performance.

[0028] Preferably, the control module is used to generate trigger time parameters corresponding to the target crab seedlings based on the capture time, conveyor belt running speed, and blowing position, and to include the trigger time parameters in the blowing trigger command; and the trigger delay parameters include total delay and single-lens delay to calibrate the timing alignment of the recognition output and the blowing action.

[0029] The control module generates trigger time parameters corresponding to the target crab seedlings based on the capture time, conveyor belt speed, and air blowing location. These trigger time parameters are included in the air blowing trigger command. Simultaneously, the trigger delay is compensated for by the total delay and single-mirror delay to calibrate the timing alignment between the recognition output and the air blowing action. This establishes a correspondence between "a certain identified crab seedling" and "the time point when it arrives at the air blowing area," offsetting the impact of link delays in acquisition, transmission, recognition, and execution response on the hit rate. This improves the timing accuracy and consistency of the air blowing trigger, avoids mis-blowing or missed blowing leading to missorting, and enhances the feasibility and stable operation of the system.

[0030] Preferably, it also includes a discharge module, which includes a first discharge bin and a second discharge bin that are separated from each other, and the position of the partition between the first discharge bin and the second discharge bin is adjustable; when the target sex is determined, the control module controls the air blowing module to blow air onto the target crab seedling to change its landing point and make it enter the first discharge bin; when the target sex is determined, the control module controls the air blowing module not to blow air and makes the crab seedling fall into the second discharge bin along its original movement trajectory, thereby realizing the separation of males and females.

[0031] By setting up a discharge module with adjustable positions, including a first discharge bin and a second discharge bin, and by controlling the airflow to change the landing point and enter the first discharge bin when the sex is determined to be the target sex, and to stop blowing air and let the air fall into the second discharge bin along the original trajectory when the sex is determined to be non-target sex, the trajectory difference corresponding to "blowing / not blowing" is transformed into clear spatial bin collection, realizing physical isolation of males and females and continuous collection. The adjustable position of the discharge bins allows the system to calibrate the landing point according to the size of the crab seedlings, the speed of the conveyor belt, and the magnitude of airflow deflection, which facilitates quick adaptation to different working conditions and reduces the risk of mixed bins, improving the fault tolerance of sorting and the adaptability to practical applications.

[0032] Preferably, the control module is used to set the opening time of the solenoid valve to define the blowing duration; the blowing module selects to open one or more air nozzles according to the coverage range of the target crab seedlings so that the airflow covers the flight path of the target crab seedlings; the control system also includes an air pressure monitoring module and a human-machine interaction module, the air pressure monitoring module is used to detect the air supply pressure status and output an alarm and / or prohibit blowing trigger when the air pressure is abnormal, the human-machine interaction module is used to set or switch control at least the following parameters: upper supplementary light on / off, lower supplementary light on / off, valve power supply on / off, air pressure detection switch, start-up valve switch, sorting mode selection, and production adjustment and motor start / stop and speed adjustment.

[0033] The control module defines the blowing duration by setting the opening time of the solenoid valve. The blowing module selects to open one or more air nozzles to cover the flight path according to the target crab seedling coverage area. This allows for controllable adjustment of the airflow amount and range under different conveying speeds and landing point requirements, improving the consistency and hit rate of trajectory deflection. At the same time, an air pressure monitoring module detects the air supply pressure and alarms and / or prohibits blowing triggering when abnormal, which can prevent deflection failure and misclassification caused by insufficient or fluctuating air pressure, thus improving the safety and reliability of system operation. A human-machine interface module provides setting and switching control of parameters such as light source, valve power supply, air pressure detection, valve testing upon startup, sorting mode, and output / motor adjustment, which facilitates on-site debugging, maintenance, and working condition switching, reduces the difficulty of operation and maintenance, and improves equipment adaptability and operability.

[0034] The technical solution of the present invention may include the following beneficial effects:

[0035] By utilizing the suspended conveyor belt to create a stable capture window, and combining simultaneous image acquisition from both upper and lower perspectives with automatic recognition and discrimination, online continuous sorting of crab seedlings can be achieved, significantly reducing reliance on manual sorting and meeting the needs of large-scale processing of large quantities of crab seedlings.

[0036] By acquiring images of the same crab larvae from both the top and bottom perspectives, the impact of random orientation of the upper and lower surfaces of the larvae, posture flipping, and local occlusion on recognition is reduced. Stable sex determination results can still be output even in the stage of crab larvae where sex characteristics are subtle, thereby improving sorting consistency and reliability.

[0037] A high-speed pneumatic solenoid valve is used to deflect the target crab seedlings in a non-contact manner, and an adjustable feeding bin is used to achieve controllable diversion of the landing point. Attached Figure Description

[0038] Figure 1 This is a flowchart of the system modules in Embodiment 1 of the present invention, which is a crab seedling sex sorting control system based on multi-view image recognition.

[0039] Figure 2This is a schematic diagram of the abdominal ROI for sex identification in a crab larvae sex separation control system based on multi-view image recognition according to the present invention.

[0040] Figure 3 This is a schematic diagram of the main interface of the industrial control screen of a crab seedling sex separation control system based on multi-view image recognition according to the present invention.

[0041] Figure 4 This is a schematic diagram of the light source setting interface of a crab larvae sex separation control system based on multi-view image recognition according to the present invention.

[0042] Figure 5 This is a schematic diagram of the camera allocation interface in a crab seedling sex separation control system based on multi-view image recognition according to the present invention.

[0043] Figure 6 This is a schematic diagram of the delay function and valve opening time setting interface in a crab seedling sex separation control system based on multi-view image recognition according to the present invention.

[0044] Figure 7 This is a three-dimensional structural diagram of a crab seedling sex separation control system based on multi-view image recognition according to the present invention.

[0045] In the diagram, 1: frame, 2: feeding module, 3: material conveying module, 4: image acquisition module, 5: image recognition module, 6: control module, 7: air blowing module, and 8: discharge module. Detailed Implementation

[0046] Example 1

[0047] In this embodiment, as Figure 1 As shown, this system consists of a feeding module, a conveying module, an image acquisition module, an image recognition module, a control module, an air blowing module, and a discharging module. The control module performs unified timing scheduling and I / O linkage. The control module can be implemented using a combination of an industrial PC and digital I / O boards (or a PLC and an industrial PC). The industrial PC is responsible for image data reception, recognition, and task queue management, while the I / O boards are responsible for outputting control signals to actuators such as speed-regulating motor drivers, supplementary lighting power supplies, and high-speed pneumatic solenoid valves. The industrial PC can communicate with the image acquisition device via USB 3.0 / Gigabit Ethernet, and with the valve drive board via 24V digital or high-speed pulse output. The industrial PC can communicate with the frequency converter / servo driver via analog 0–10V or serial communication to set speed commands. All communication and execution signals are uniformly timestamped within the control module for calibration and traceability.

[0048] In one optional implementation, the feeding module includes a feeding hopper, a baffle plate, and a vibrating feeder / vibrating screen. The baffle plate moves along a slide rail to change the effective opening of the discharge port, thereby achieving a continuous and controllable feeding rate. The vibration frequency and amplitude of the vibrating feeder can be set to preset levels, so that the crab seedlings are loosened and dispersed before entering the conveyor belt, and fall in a single layer or near-single layer as much as possible. To further suppress the stacking of crab seedlings on the conveyor belt surface, a height limit bar or flexible scraper can be installed above the conveyor belt to disperse the stacked individuals into a single layer when passing through the height limit section. At the same time, guide baffles can be installed on both sides of the conveyor belt to limit lateral drift and form a more stable off-belt queue. The conveying module uses a speed-regulating motor to drive the conveyor belt, and the control module controls the output on the interface (…). Figure 3 The entrance provides motor start / stop and speed adjustment, and realizes the linkage between belt speed and unloading according to the preset feeding density and belt speed strategy during operation: when the number of targets in the belt area per unit time is too large or the occlusion ratio increases, the control module can reduce the opening of the insert plate / reduce the amount of vibrator deployment or increase the belt speed, so that the individuals entering the air capture window tend to be more dispersed.

[0049] like Figure 4 The image acquisition module is used to acquire top and bottom view images of the same crab seedling during its aerial movement, and can optionally be configured with top and bottom supplementary lighting sources.

[0050] In terms of structural layout, the upper and lower image acquisition devices are fixed inside the upper and lower sorting boxes, with the optical axis aligned with the same spatial capture window after leaving the sorting zone. To ensure the correspondence between the upper and lower viewing angles of the same crab larva, the control module adopts a synchronous triggering strategy in one implementation: when the crab larva reaches the preset trigger position instantaneously after leaving the sorting zone, the same trigger signal simultaneously triggers the exposure of the upper and lower image acquisition devices; optionally, the trigger signal comes from the photoelectric sensor, encoder counting threshold, or visual front-end detection signal upstream of the sorting zone, and a unified trigger sequence number ID is generated upon triggering. To adapt to the short time window of the airborne movement, the supplementary lighting source can be triggered synchronously with the camera exposure and adopt a short pulse high-brightness mode, so that the image still has sufficient contrast and clarity under short exposure.

[0051] The image recognition module is used to distinguish between males and females and output the results. In this embodiment, the distinction between males and females is based on the morphological differences in the pelvic fin region of the crab larvae; for example... Figure 2The diagram shows the ROI (Region of Interest) for abdominal recognition (blue and red boxes indicate the abdominal ROIs of two crab larvae, respectively). The image recognition module first performs target localization on the top-view and / or bottom-view images. Target localization can be achieved by: obtaining the maximum connected component through foreground segmentation based on grayscale thresholds, or by using a lightweight object detection network to output the bounding box of the crab larvae. After obtaining the bounding box of the crab larvae, the abdominal ROI is generated based on the relative position of the geometric center of the bounding box and the abdomen. The ROI can be cropped according to a preset ratio (e.g., taking a rectangular window proportionally in the lower half or lower middle part of the bounding box), and the ROI is geometrically normalized (scaled to a uniform size) for subsequent feature calculation. Subsequently, pelvic fin extraction and quantization are performed within the ROI: After brightness normalization, noise reduction, and edge enhancement of the ROI, adaptive threshold segmentation is used to obtain candidate pelvic fin regions. Then, morphological opening and closing operations are performed to remove interference such as leg filaments, retaining the main connected domain of the pelvic fin. Contours are extracted from the main connected domain, and preset geometric indices are calculated to form feature vectors. The geometric indices include at least the pelvic fin opening width, pelvic fin length, pelvic fin area, and the number of contour curvature peaks. Sex determination can be performed using a sample database matching method or a classification model method: When using sample database matching, multi-view samples of female / male crab larvae are collected and labeled in advance to establish a sample database. The current ROI is matched with the sample database for similarity and the matching degree (0%~100%) is output. When the matching degree meets the preset threshold, the female / male result is output. When using the classification model method, the feature vector or ROI image is input into the classifier to output the female / male probability and the discrimination result is output according to the threshold. For multi-view fusion, under the same trigger sequence ID, the upper and lower views output the discrimination results separately and then fuse them: if they are consistent, the consistent result is output directly; if they are inconsistent, the one with higher confidence is output, or the final result is output by voting according to preset weights. The confidence can be determined by indicators such as the matching degree, the integrity of the main connected component of the ventral fin, and the edge sharpness. To avoid mismatch caused by multiple crab larvae entering the field of view at the same time, the image recognition module can optionally generate ROI only for the nearest target corresponding to the trigger sequence ID (e.g., the target closest to the center of the window) during the target localization stage, and mark the trigger sequence number as an invalid sample and do not trigger the blowing when multiple targets are detected in the same frame and cannot be stably associated.

[0052] The control module maps the recognition output to a blowing action, ensuring consistency between the recognition object and the execution object. In one implementation, the control module generates a task record for each trigger. The task record includes at least: trigger sequence number (ID), capture timestamp, recognition result, the target's lateral position (which can be calculated from image coordinates), and the planned blowing channel number. The control module maintains a task queue and schedules execution according to the trigger sequence number (ID), avoiding high-concurrency timing disruptions. The trigger time parameter can be generated using a time window compensation method: a base delay is determined based on the capture timestamp and structural calibration parameters, then a total delay and camera delay (single-lens delay) compensation term are added to align the recognition output with the blowing action. The delay parameter... Figure 6 The delay function is set in the interface shown, and supports clearing and resetting; the valve opening time is also used as the blowing duration. Figure 6 Interface settings.

[0053] Regarding the selection of air nozzle channels, the air blowing module can be configured as a single nozzle or a multi-nozzle array. The control module selects and activates the corresponding nozzle based on a calibration mapping table between the target's lateral position in the image and the nozzle number in the image coordinates. When the target coverage area is large or the landing point boundary requires greater stability, one or more adjacent nozzles can be activated simultaneously to achieve coverage. To ensure mapping accuracy, a one-time calibration can be performed during the installation and debugging phase: a calibration ruler or calibration board is placed within the capture window to establish the conversion relationship between pixel coordinates and physical coordinates, and the position of the injection centerline of each nozzle in the physical coordinate system is recorded, thereby forming a mapping table for the control module to use.

[0054] The discharge module includes a first receiving bin and a second receiving bin, and the position of the partition between the two bins is adjustable. Optionally, an isolation baffle or guide baffle is set between the two bins and its position is adjustable to limit the landing point boundary and for on-site calibration. During implementation, the control module sets the conveyor belt speed to a low level, so that the initial horizontal velocity of crab larvae that have not been blown off the belt is small and their falling trajectory is mainly vertical. Under this condition, female crabs, which correspond to the non-blown category, fall into the first receiving bin after leaving the belt in a near-free fall. At the same time, the blowing direction is adjusted from top to bottom to forward: the air nozzle is installed diagonally below the falling trajectory of the crab larvae, and the spray direction is towards the second receiving bin. When the image recognition module determines that it is a male crab, the control module opens the solenoid valve when the trigger time is reached, so that the male crab is displaced forward by the diagonally downward forward airflow during the fall and enters the second receiving bin. To achieve stable compartmentalization, the relative positions of the first and second receiving compartments, the positions of the isolation / guide baffles, and the air nozzle elevation angle are calibrated during the installation and commissioning phase. This ensures that the free fall landing area and the blown landing area stably enter the corresponding receiving compartments. An air compressor and air tank can be configured on the air supply side, and a pressure detection switch and valve power switch are provided in the system settings interface. Figure 4 This is to support pre-run checks and anomaly management.

[0055] In this embodiment, the female crabs are assigned to a non-air-blowing path to reduce the risk of force disturbance and collision during the sorting process. In aquaculture production scenarios, female crabs usually have higher economic value, so a gentler feeding path is preferred for female crabs to improve survival and marketability.

[0056] Optionally, the control module executes a start-up valve testing procedure during the start-up phase, detecting valve response and gas path connectivity through a short-term valve opening, and prohibiting entry into the intelligent sorting operation state if the detection fails. Figure 4 (Initial start-up valve switch item).

[0057] The system's human-computer interaction is completed through an industrial control screen: the main interface is used to display the entry point and operating status. Figure 3 The system settings are used for configuring switches for light sources, valve power supplies, air pressure detection, and valve start-up testing. Figure 4 Camera allocation is used for configuring parameters such as structure type, number of camera channels, and nozzle density. Figure 5 The delay function is used to set the valve opening time and delay compensation parameters. Figure 6 ).

[0058] like Figure 7 The system in this embodiment includes: a frame 1, a feeding module 2, and a conveying module 3;

[0059] The image acquisition module 4, image recognition module 5, and control module 6 are integrated into the same integrated housing, which is fixedly mounted on the frame 1. The integrated housing contains a camera and its interface circuit for image acquisition, a processing unit for image recognition, and a control unit for outputting control signals.

[0060] It also includes an air blowing module and a material discharge module.

[0061] Two diversion methods are employed: First, in the preferred scheme, female crabs fall freely into the rear compartment, while male crabs are blown forward into the front compartment from below by air valve nozzles. This reduces the risk of stress and damage from collisions and mechanical contact, improving survival rates and protecting high-value individuals. Second, in the preferred scheme, the conveyor belt speed is appropriately increased. Female crabs fall freely into the front compartment along a parabolic trajectory, while the identified male crabs are affected by the downward airflow from the air valve nozzles above, altering their original parabolic trajectory and falling downwards into the rear compartment.

[0062] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

[0063] Example 2

[0064] In Example 1, the control module calculates the air-blowing trigger time based on the capture time and the conveyor belt speed, and aligns the timing of the recognition output and the air-blowing action by setting the total delay and camera delay. Simultaneously, the air-blowing duration is typically set with a fixed valve opening time and manually fine-tuned to calibrate the landing point. While this implementation can achieve basic sorting, the fixed delay and fixed valve opening parameters can easily lead to landing point deviation and reduced flow stability when operating conditions change due to individual differences in crab larvae, posture fluctuations, air supply fluctuations, and conveyor belt speed adjustments. Furthermore, frequent manual adjustments are required. To reduce reliance on empirical tuning and improve the repeatability and calibrability of sorting control, this embodiment improves the triggering timing and blowing control strategy of the control module without changing the overall structural composition of Embodiment 1. This allows the control module to comprehensively consider physical constraints such as separation height, initial horizontal velocity, flight time, and airflow impulse, calculate the triggering time and valve opening time that match the target crab seedlings, and output executable control commands for nozzle selection and valve opening duration, thereby achieving stable flow separation within the limits allowed by the on-site structure and air supply conditions.

[0065] In this embodiment, the sorting process includes the following steps:

[0066] S1. The feeding module feeds and loosens the crab seedlings, allowing them to enter the conveyor belt of the transport module in a dispersed state.

[0067] S2. The material conveying module controls the operation of the conveyor belt, transporting the crab larvae to the departure point at the end of the conveyor belt, causing the crab larvae to leave the conveyor belt and enter the aerial capture area. The control module records or acquires the linear speed of the conveyor belt, which can be calculated from the set value of the speed-regulating motor or measured and read by the encoder.

[0068] S3. The image acquisition module acquires top-view and bottom-view images of the same crab larva during its aerial movement and sends them to the image recognition module. Based on the top-view or bottom-view image, the image recognition module outputs a female / male crab identification result and returns the identification result along with the capture timestamp to the control module. The control module establishes a task record for each capture target. The task record includes at least: a capture timestamp, a female / male identification result, and the target's lateral position coordinates in the image coordinate system for nozzle selection.

[0069] S4. Calculate the reference set of natural landing points based on the natural flight model;

[0070] In this step, the control module introduces the takeoff height h and the equivalent damping parameter κ to model the natural flight during takeoff, in order to obtain the natural landing point reference set X. The takeoff height h is the effective drop from the takeoff point to the receiving plane, which can be obtained by measuring the structural dimensions and stored as a system configuration parameter. The equivalent damping parameter κ is used to characterize the comprehensive drag effect of the crab larvae's horizontal movement during the takeoff phase; it can be pre-calibrated and stored according to the crab larvae's specifications, or statistically analyzed during on-site trial operation and written into the configuration file.

[0071] The control module uses the initial horizontal velocity v at the instant of separation from the belt. x (0)=v b Establish a linear damping model for horizontal motion:

[0072]

[0073] Among them, v x Indicates the horizontal velocity of the crab larvae along the transport direction during the airborne phase; symbol Indicates v x The derivative with respect to time; t represents the time variable during the takeoff phase;

[0074] Therefore, we can conclude that:

[0075]

[0076] Where e is the base of the exponential function of the natural constant; v b This represents the initial value of the horizontal velocity at the instant of separation from the belt.

[0077] The control module approximates the vertical takeoff time as the gravity-dominated descent time:

[0078]

[0079] Among them, t f represents the time of flight from the point of departure to the receiving surface; g represents the gravitational acceleration constant.

[0080] The control module integrates the horizontal displacement to obtain the natural horizontal displacement:

[0081]

[0082] Where x0 represents the natural horizontal displacement of the target crab larvae from the departure point to the ground under conditions of no air blowing; integral This indicates the time interval from 0 to t. f The horizontal velocity is integrated to obtain the displacement.

[0083] And define the natural flight output as the natural landing point reference set:

[0084]

[0085] Where X represents the set of natural landing points; Δ xunc This represents the uncertain range of the natural landing point, used to characterize the range of landing point fluctuations caused by attitude flipping and positioning errors.

[0086] Δx unc It can be pre-statistically counted and stored according to specification levels, or calculated based on image pose fluctuation index, for example:

[0087]

[0088] Where α represents the calibration coefficient (dimensionless or set according to unit consistency); σ θ The statistical measure representing the attitude angle fluctuation (e.g., attitude angle standard deviation) is θ; θ represents the attitude angle variable. Using set X, the control module obtains the natural landing position and fluctuation range of the target crab larvae under non-inflation conditions, providing a benchmark for subsequent inflation deflection calculations and landing window determination.

[0089] S5. Calculate the blowing deflection set Y based on the airflow impulse model;

[0090] In this step, the control module introduces the airflow impulse coefficient J and the equivalent mass m of the target crab seedling to model the airflow deflection, so as to obtain the airflow deflection set Y. Here, J represents the airflow impulse coefficient; m represents the equivalent mass of the target crab seedling, which can be preset and stored according to the specification level or obtained from the sample weighing statistics.

[0091] The control module represents the equivalent impulse within the valve opening time τ as:

[0092]

[0093] Where I represents the equivalent impulse; τ represents the valve opening time of the solenoid valve, i.e. the duration of air blowing.

[0094] The speed increment obtained by the target crab seedlings is:

[0095]

[0096] Where Δv represents the forward velocity increment introduced by the blowing action.

[0097] The control module pre-stores the structural position parameters of the nozzle line of action relative to the off-zone point, and determines the flight time t from the off-zone to the nozzle line of action. jet The t jet It can be obtained by converting the structural distance to the average horizontal speed or by calibration during trial operation. Wherein, t jet This indicates the time it takes for the target crab larvae to move from the point of departure to the effective action line of the nozzle.

[0098] The control module uses this information to determine the remaining time from the start of the blowing action to landing:

[0099]

[0100] Among them, t r This indicates the remaining flight time to the receiving surface after the nozzle has been applied.

[0101] This results in the additional horizontal displacement caused by blowing air:

[0102]

[0103] Where, Δx J This indicates the additional horizontal displacement caused by the blowing air; the subscript J indicates that the displacement is calculated primarily by the airflow impulse coefficient J.

[0104] And define the blowing deflection output as:

[0105]

[0106] Where Y represents the set of blowing deflections; This indicates the recommended valve opening time that satisfies the second receiving bin landing point window constraint and is selected according to the criteria.

[0107] S6: Merge the natural landing point reference set and the blowing deflection set to generate a control command set Ω and execute the flow splitting;

[0108] In this step, the control module pre-stores the landing point windows corresponding to the first and second receiving bins. and ;

[0109] in, This indicates the allowable natural drop point range for the first receiving bin; and These represent the lower and upper bounds of the interval, respectively.

[0110] This indicates the permissible range of blowdown points in the second receiving bin; and These represent the lower and upper bounds of the interval, respectively. The landing point window is determined by the location of the receiving hopper and the position of the isolation / guide baffle, and can be calibrated during the commissioning phase. The control module uses the natural landing point x0 and the air deflection Δx as the reference points. J Generate the final control command set Ω and control the air blowing module to perform flow splitting.

[0111] When the determination result is a female crab, the control module does not output an air-blowing trigger command, allowing the female crab to fall along its natural flight trajectory to the first receiving bin, and performs window verification on the natural landing point:

[0112]

[0113] When the crab is identified as male, the control module outputs a blowing trigger command, causing the male crab to be propelled to the second receiving bin by a downward-sloping, forward-spraying airflow during its descent. The control module first determines the feasible range of the valve opening time τ using the following inequality:

[0114]

[0115] This represents the additional displacement Δx that can be produced by blowing air. J .

[0116] Therefore, we get:

[0117]

[0118] The control module operates within the aforementioned feasible range and the equipment's permissible valve opening time range [τ]. min ,τ max The recommended valve opening time τ is determined from the intersection of [the values ​​of the valve opening time and the valve opening time]. ∗ Among them, τ min With τ max These represent the minimum and maximum valve opening times allowed by the equipment, which are determined and pre-configured by the solenoid valve and the air circuit capacity.

[0119] In a preferred implementation, to reduce overblowing and improve landing stability, the control module minimizes the following expression within the feasible region to determine τ. ∗ :

[0120]

[0121] in Δx represents the value of the independent variable τ that minimizes the objective function. J (τ) represents Δx J Consider it a function of τ; x 2,mid This indicates the midpoint position of the second receiving hopper's landing point window, and:

[0122]

[0123] Here, the right side of the fraction is the arithmetic mean of the endpoints of the interval; λ represents the penalty coefficient, which is used to constrain and weigh the velocity increment term.

[0124] The control module obtains τ ∗ The trigger time t is then calculated. fire The control module represents the system link delay as:

[0125]

[0126] Among them, t lat Indicates total link delay; t imgIndicates the image acquisition and transmission delay; t infer t represents the computational latency for image recognition inference. commt Indicates control communication delay; t val This indicates the response delay of the solenoid valve; the above delay can be compensated for by the total delay of the control module and the camera delay parameters.

[0127] The trigger time is calculated based on the capture timestamp and the arrival time of the nozzle's line of action:

[0128]

[0129] Among them, t fire This indicates the timestamp corresponding to the moment the valve is opened.

[0130] Simultaneously, the control module selects nozzle number n=f(y) or nozzle group based on the target's lateral position coordinate y, so that the airflow acts on the effective falling path of the target crab larva. Here, n represents the nozzle number or nozzle group number; f(⋅) represents the mapping function from the lateral position to the nozzle number obtained from calibration.

[0131] Finally, the control module outputs and executes a set of control commands:

[0132]

[0133] Where Ω represents the set of control commands; in this embodiment, to ensure the feasibility of the above modeling calculations, the control module performs parameter consistency verification before entering the sorting mode: when it detects:

[0134]

[0135] or

[0136]

[0137] When the structural parameter configuration is abnormal, entry into the sorting process is prohibited; when the valve opening time feasible domain is empty or equal to [τ], the process is considered abnormal. min ,τ max When there is no overlap, an alarm is output and a prompt is made to adjust the conveyor belt speed, receiving bin position, or air supply pressure / nozzle configuration to restore the sortable state. The above "≤0" indicates that the remaining time is not positive, which prevents the effective deflection calculation after the nozzle action from being completed.

[0138] Through the above steps S1–S6, physical constraint parameters such as the separation height h, equivalent damping parameter κ, and airflow impulse coefficient J are introduced, enabling the control module to calculate the natural landing point reference set X and the blowing deflection set Y, and further fuse them to obtain the control command set Ω. Thus, even when the on-site working conditions change, the blowing trigger sequence and valve opening time can still be output in a calibrable manner, achieving stable diversion of female crabs falling freely into the first receiving bin and male crabs being blown obliquely downwards and forwards into the second receiving bin.

Claims

1. A crab larvae sex separation control system based on multi-view image recognition, characterized in that, include: Material handling module, image acquisition module, image recognition module, control module, and air blowing module; The material conveying module is used to control the operation of the conveyor belt, so that the crab seedlings leave the conveyor belt at the end of the conveyor belt and form an airborne motion; The image acquisition module is used to acquire top-view and bottom-view images of the same crab seedling during its aerial movement. The image recognition module is used to output the female crab identification result or the male crab identification result based on the upper-view image or the lower-view image; The control module is used to receive the sex determination result and output a blowing trigger command; The blowing module is used to control the high-speed pneumatic solenoid valve to open the nozzle and spray air according to the blowing trigger command, so that the airflow acts on the target crab seedling to change its flight trajectory, thereby achieving male and female separation.

2. The crab larvae sex separation control system based on multi-view image recognition according to claim 1, characterized in that, It also includes a feeding module, which includes a feeding hopper, and the feeding rate can be adjusted by adjusting the opening of the bottom plate of the feeding hopper; The feeding module also includes a vibrating feeder, which is located below the feeding hopper to vibrate the crab seedlings, loosen and evenly disperse them, and allow them to enter the conveyor belt in a dispersed state.

3. A crab larvae sex separation control system based on multi-view image recognition according to claim 2, characterized in that, The material conveying module controls the speed-regulating motor to adjust the speed of the conveyor belt, so that the crab seedlings are conveyed in a single layer or near-single layer on the surface of the conveyor belt and stably leave the belt to enter the aerial capture area.

4. A crab larvae sex separation control system based on multi-view image recognition according to claim 3, characterized in that, The image acquisition module includes image acquisition devices arranged vertically, used to capture images of the same crab seedling in mid-air from multiple perspectives and output the upper perspective image and the lower perspective image.

5. A crab larvae sex separation control system based on multi-view image recognition according to claim 4, characterized in that, The image acquisition module also includes an upper supplementary light source and a lower supplementary light source, which are used to cooperate with the upper image acquisition device and the lower image acquisition device to capture images during aerial capture.

6. A crab larvae sex sorting control system based on multi-view image recognition according to claim 5, characterized in that, The image recognition module uses the shape of the pelvic fins on the abdomen of crab larvae as a feature to distinguish between males and females, and fuses the recognition results from the upper and lower viewpoints to output the final discrimination result.

7. A crab larvae sex separation control system based on multi-view image recognition according to claim 6, characterized in that, The image recognition module matches the recognition results with a pre-stored multi-view sample database to output a sex determination result.

8. A crab larvae sex separation control system based on multi-view image recognition according to claim 7, characterized in that, The control module is used to generate trigger time parameters corresponding to the target crab seedlings based on the capture time, conveyor belt speed, and blowing position, and to include the trigger time parameters in the blowing trigger command; and the trigger delay parameters include total delay and single-lens delay to calibrate the timing alignment between the recognition output and the blowing action.

9. A crab larvae sex separation control system based on multi-view image recognition according to claim 8, characterized in that, It also includes a discharge module, which includes a first discharge bin and a second discharge bin that are set apart from each other, and the positions of the first discharge bin and the second discharge bin are adjustable; when the target sex is determined, the control module controls the air blowing module to blow air onto the target crab seedling to change its landing point and make it enter the first discharge bin; when the target sex is determined, the control module controls the air blowing module not to blow air and makes the crab seedling fall into the second discharge bin along its original movement trajectory, thereby realizing the separation of males and females.

10. A crab larvae sex separation control system based on multi-view image recognition according to claim 8, characterized in that, The control module is used to set the opening time of the solenoid valve to define the blowing duration; the blowing module selects to open one or more air nozzles according to the coverage range of the target crab seedlings so that the airflow covers the flight path of the target crab seedlings; the control system also includes an air pressure monitoring module and a human-machine interaction module. The air pressure monitoring module is used to detect the air supply pressure status and output an alarm and / or prohibit blowing trigger when the air pressure is abnormal. The human-machine interaction module is used to set or switch control at least the following parameters: upper supplementary light on / off, lower supplementary light on / off, valve power on / off, air pressure detection switch, start-up valve switch, sorting mode selection, and production adjustment and motor start / stop and speed adjustment.

Images