CN-121730234-B - Meat fish homogeneous phase feeding intelligent isolation device based on machine vision and early warning method

CN121730234BCN 121730234 BCN121730234 BCN 121730234BCN-121730234-B

Abstract

The invention relates to the field of fish culture early warning, in particular to an intelligent isolation device and an early warning method for meat fish homogeneous phase feeding based on machine vision. The device mainly comprises a culture box, a vision acquisition module, a system control and data processing mechanism and an array type air curtain generating mechanism positioned at the bottom of the box, wherein the vision acquisition module acquires a water body image in the culture box in real time, an embedded terminal operates a gesture sensing network capable of simultaneously outputting a target space position and a plurality of structural key points, vector included angles and Euclidean distances of a spine of the fish body are calculated in real time by extracting anatomical key points such as a head part, a dorsal fin and a caudal peduncle of the fish body, when the target fish body is monitored to present a high-curvature attack gesture, the system immediately locks a conflict area through coordinate mapping, drives an integrated microporous aeration pipe array to release compressed air, and generates a vertical non-contact type air bubble barrier between an attacker and an attacked person, so that physical isolation is realized. The invention is suitable for fish culture.

Inventors

XIA HAORAN
PANG TAO
WU JIXIAO
CHEN XIAOYAN
Zha Songyou

Assignees

四川农业大学

Dates

Publication Date: 20260508
Application Date: 20260226

Claims (7)

1. The meat fish homogeneous phase feeding intelligent isolation device based on machine vision is characterized by comprising a culture box (3), a vision acquisition module, a system control and data processing mechanism and an array type air curtain generation mechanism, wherein the system control and data processing mechanism is a chassis embedded edge computing terminal (1), the culture box (3) is positioned above the chassis embedded edge computing terminal (1), the vision acquisition module consists of a suspension type main camera module (8) and an auxiliary camera module (6), the suspension type main camera module (8) is connected with support frames (4) on two sides through flexible flat cables (7) on two sides, the auxiliary camera module (6) is arranged on four sides of the inner wall of the culture box (3) respectively, and the array type air curtain generation mechanism consists of an integrated microporous aerator array (5), is controlled by electromagnetic valves and is regularly distributed at the bottom of the culture box (3); The method comprises the steps of acquiring water video flow in a culture box in real time through a vision acquisition module consisting of a hanging type main camera shooting module and an auxiliary camera shooting module, monitoring the pose of all fishes, extracting the head, the body center, the dorsal fin and the tail anatomical points of the fishes through a pose sensing network running in an embedded edge computing terminal, calculating the vector included angle and Euclidean distance of the spinal column of the fishes in real time, and when the embedded edge computing terminal monitors that the target fishes present a high curvature attack pose, namely similar feeding events occur, locking a conflict area through coordinate mapping immediately by a system control and data processing mechanism, driving an integrated microporous aeration pipe array at the bottom of the culture box to respond in millisecond time, releasing compressed air, generating a vertical non-contact bubble barrier between an attacker and the attacked person, and realizing physical blocking.
2. The intelligent isolation device for meat fish homogeneous phase feeding based on machine vision according to claim 1, wherein guide rail assemblies (2) are installed on the upper portion of the chassis embedded edge computing terminal (1) and distributed on two sides of the cultivation box (3), the guide rail assemblies (2) are composed of guide rails (201) and limiting blocks (202), and the supporting frame (4) is in contact with the guide rails (201) through pulleys in holes in the bottom.
3. The intelligent isolation device for meat fish homogeneous phase feeding based on machine vision according to claim 1, wherein the suspension type main camera module (8) consists of a nacelle (801) and a main camera (802), the nacelle (801) is connected with flexible flat cables (7) on two sides and is connected with rotating wheels in two side supporting frames (4), and the main camera (802) is positioned below the nacelle (801).
4. The intelligent isolation device for homogeneous phase feeding of carnivorous fishes based on machine vision according to claim 1, wherein the integrated microporous aerator pipe array (5) is divided into a four-hole microporous aerator pipe (501), a three-hole microporous aerator pipe (502) and a single-hole microporous aerator pipe (503) according to different positions, and is driven by electromagnetic valves.
5. The intelligent pre-warning method for the similar predation of the carnivorous fishes based on the machine vision is applied to the intelligent isolation device for the similar predation of the carnivorous fishes based on the machine vision as claimed in any one of claims 1 to 4, and is characterized in that the pre-warning method comprises the following steps: S1, preprocessing a video stream in a culture box acquired by a real acquisition module, removing salt and pepper noise caused by water bubbles and suspended particles by adopting Gaussian filtering, and enhancing image contrast by adopting self-adaptive histogram equalization; s2, inputting the preprocessed image into a gesture sensing network, and extracting anatomical points of the head, the body center, the dorsal fin and the tail of the fish body; The gesture sensing network adopts a pre-trained YOLOv-Pose neural network model, the YOLOv-Pose neural network model adopts a layered visual sensing architecture, wherein a backbone network adopts a CSPDARKNET structure to perform feature extraction, RGB pixel values are changed into overall structural features of fish body outline, fish body texture, fish head and tail morphological differences and bending gesture layer by layer, and a head network is configured to output two types of information of each detection target: first type bounding box information: ; Wherein B represents the bounding box itself, Representing the x-axis coordinates of the center point of the bounding box, Representing the y-axis coordinates of the center point of the bounding box, The width of the bounding box is indicated, Representing the height of the bounding box; the bounding box is used for limiting the space range of the target fish body and is used as an object-level constraint condition for point location extraction, gesture analysis and behavior judgment; Second, fish body point location coordinates: identifying and recording the head space coordinates, body center or dorsal fin space coordinates and tail space coordinates of the fish skeleton; Head coordinates: ( ); Body center or dorsal fin coordinates: ( ); Tail coordinates: ( ); s3, calculating the spine bending angle according to the point position coordinates of each detected fish body target; Firstly, calculating corresponding vectors according to the coordinates of the point positions of the fish body; body center or dorsal fin to head direction vector: ; body center or dorsal fin to tail direction vector: ; calculating the spine bending angle by using vector dot product formula : , ; S4, if the spine bending angle in a certain individual continuous multi-frame Less than a threshold value Finding a fish individual nearest to the individual, recording the coordinate of the fish individual nearest to the fish which presents the attack accumulation state, and calculating the Euclidean distance: ; in the formula, The distance of the euro type is expressed, , Representing the coordinates of the individual presenting the offensive power status, , Coordinates representing the fish individual nearest to the former; s5, defining an attack judgment logic function : ; In the formula, Representing a preset Euclidean distance threshold; If it is 0, Judging that the possible similar food-like events do not occur; If it is Then it is determined that a possible similar food event has occurred.
6. The intelligent pre-warning method for similar feeding of carnivorous fishes based on machine vision according to claim 5, which is characterized in that the pre-warning method further comprises: When the possible similar food events are judged to occur, calculating coordinate mapping of the occurrence of the events: ; in the formula, For the image coordinates in the video stream captured in the camera, The real coordinates are corresponding to the integrated microporous aerator pipe array coordinates; 。
7. the intelligent pre-warning method for similar feeding of carnivorous fishes based on machine vision according to claim 5, which is characterized in that the pre-warning method further comprises: when the possible similar food events are judged, the system control and data processing mechanism sends instructions to control the integrated microporous aeration pipe array at the bottom of the cultivation box to release compressed air, and a vertical non-contact bubble barrier is generated between an attacker and an attacked person, so that physical blocking is realized.

Description

Meat fish homogeneous phase feeding intelligent isolation device based on machine vision and early warning method Technical Field The invention relates to the field of fish culture early warning, in particular to an intelligent isolation device and an early warning method for meat fish homogeneous phase feeding based on machine vision. Background With the development of the aquaculture industry to the intensive and industrial direction, high-density circulating water culture has become a mainstream mode. In the breeding stage of fries of high economic value carnivorous fishes such as groupers, mandarin fish, micropterus salmoides, and the like, the same type of feeding phenomenon is easy to occur due to natural aggression and territory consciousness. Research shows that the loss rate caused by the same kind of food can reach 30% -50% in the metamorphosis period and the growth difference period of the seedlings, and the cultivation benefit is severely restricted. In order to solve this problem, the existing technical solutions are mainly divided into two categories: The first type is periodic mechanical classification. The fries with different specifications are separated by manual net pulling or screen devices at regular intervals. However, the method has the obvious defects that firstly, the method has hysteresis, cannot be intervened at the moment of attack, and can only be remedied after the moment, secondly, the mechanical screening belongs to contact operation, mechanical damage on the surface of the fish body is extremely easy to cause, such as scale falling off, so that secondary infection such as saprolegniasis is caused, and thirdly, the frequent artificial interference can cause strong stress reaction of the fish shoals to influence ingestion and growth. The second category is based on traditional video monitoring early warning systems. At present, monitoring equipment aiming at underwater environment is available in the market, and although the number of fish shoals or approximate moving tracks can be identified by utilizing an image processing technology, the method has the following defects that most algorithms only stay on a monitoring layer and lack linkage with an executing mechanism to realize closed loop control, secondly, the existing behavior identification algorithm is mainly based on a simple movement speed threshold value, is difficult to distinguish normal quick tour and accumulation posture before attack, has higher false alarm rate, and finally, lacks an effective non-contact blocking means, even if attack behaviors are identified, the attack behaviors can be manually processed only through alarm prompt, and the millisecond blocking requirement can not be met. Fish eyes have evolved a fixed spherical lens and wide angle field of vision to accommodate underwater vision, which makes it difficult to quickly focus and filter motion blur. The bubble curtain can generate dense and irregular air-water interface, and forms an optical noise barrier which continuously shakes and breaks in water. The continuous light scattering and image distortion just attack the natural short plate of fish eye vision, so that the natural short plate cannot stably acquire clear images, further cause sensory interference and behavioral obstruction, and can influence predation or communication with the same kind. In summary, an intelligent device capable of accurately recognizing the pre-attack gesture by utilizing machine vision and realizing millisecond-level instantaneous blocking by adopting a non-contact mode such as pneumatic interference is developed, and the intelligent device has important application value for improving the survival rate of carnivorous fish fries and animal welfare. Disclosure of Invention The invention aims to overcome the defects of the prior art, and provides an intelligent isolation device and an early warning method for similar and mutually feeding of carnivorous fishes based on machine vision, which solve the problem of pain points of mutual killing of rare carnivorous fish fries in high-density cultivation, avoid the stress damage of fish bodies caused by traditional mechanical driving, and remarkably improve the survival rate of raising seedlings and the intelligent level of cultivation management. The invention adopts the following technical scheme to achieve the aim, and in a first aspect, the invention provides an intelligent isolation device for meat fish homogeneous phase feeding based on machine vision, which comprises the following components: The system comprises a cultivation box 3, a vision acquisition module, a system control and data processing mechanism and an array type air curtain generating mechanism, wherein the system control and data processing mechanism is a chassis embedded edge computing terminal 1, the cultivation box 3 is positioned above the chassis embedded edge computing terminal 1, the vision acquisition module consists of a hanging type main camer