CN-122023900-A - Mutton sheep ectoparasite intelligent recognition system based on machine vision

CN122023900ACN 122023900 ACN122023900 ACN 122023900ACN-122023900-A

Abstract

The invention discloses an intelligent recognition system for mutton sheep ectoparasites based on machine vision, which comprises an image acquisition module, an image preprocessing and enhancing module and a processing module, wherein the image acquisition module is used for acquiring original visual data of mutton sheep body surfaces, the image preprocessing and enhancing module is connected to the image acquisition module and is used for carrying out standardization processing on original images and enhancing micro-object-oriented areas, and the image preprocessing and enhancing module comprises a self-adaptive interested area extraction sub-module and a detail enhancing flow. The intelligent recognition system for the ectoparasites of the mutton sheep based on the machine vision systematically solves the fundamental problem of weakening of the characteristics of the tiny targets under the complex hair background by introducing a detail enhancement mechanism of self-adaptive interested region extraction and multi-scale image fusion. According to the technical scheme, the whole image is not subjected to equal processing, but is intelligently focused on a suspected region, and microcosmic details of the suspected region are enhanced by adopting an algorithm, so that the omission ratio is reduced at the source.

Inventors

ZHU GUANGQIN
WANG BING
CHI LAN
SHI FENGYUN
SUI SHAOPU

Assignees

徐州生物工程职业技术学院

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (8)

1. The intelligent recognition system for the ectoparasites of the mutton sheep based on the machine vision is characterized by comprising an image acquisition module, a recognition module and a recognition module, wherein the image acquisition module is used for acquiring original visual data of the body surface of the mutton sheep; The image preprocessing and enhancing module is connected to the image acquisition module and is used for carrying out standardization processing on an original image and carrying out region enhancement facing to a tiny target, the image preprocessing and enhancing module comprises an adaptive region of interest extraction submodule and a detail enhancing flow, the adaptive region of interest extraction submodule is used for initially positioning regions with dense hair textures or abnormal color plaques in the image based on a pre-trained shallow neural network and judging the regions as potential parasite habitat areas, and the detail enhancing flow is used for processing each extracted potential habitat image by adopting a multi-scale image fusion algorithm based on a Laplace pyramid so as to generate a high-definition focusing sub-image; The multi-scale feature fusion detection module is connected to the image preprocessing and enhancing module; The life cycle classification module is connected to the multi-scale feature fusion detection module; the decision and output module is connected with the life cycle classification module and the multi-scale feature fusion detection module.
2. The intelligent recognition system of the meat sheep ectoparasites according to claim 1, wherein the multi-scale feature fusion detection module is used for recognizing and positioning parasite targets from the focusing sub-images, and the core of the multi-scale feature fusion detection module is a hierarchical convolutional neural network model specially designed for tiny target detection, and the model adopts an encoder-decoder architecture and integrates a dual attention mechanism, wherein the dual attention mechanism comprises a spatial attention sub-module and a channel attention sub-module.
3. The intelligent recognition system of the ectoparasite of the mutton sheep based on machine vision, wherein the life cycle classification module is used for classifying the detected parasite targets in a growth stage, a fine-granularity image classification network is built in the life cycle classification module, the network takes a lightweight convolutional neural network as a backbone, and a multi-task learning head is connected to the tail end of the network, and the multi-task learning head simultaneously executes a main classification task for classifying the targets into main life cycle stages and an auxiliary classification task for further distinguishing states of the targets in adult stages.
4. The intelligent recognition system of the ectoparasites of the mutton sheep based on machine vision according to claim 1, wherein the decision and output module is used for integrating detection and classification results and generating management decisions and visual reports, and a rule engine is arranged in the decision and output module, and the rule engine is pre-stored with risk levels and processing suggestion mapping tables corresponding to different parasite types and different life cycle stages.
5. The intelligent recognition system for the ectoparasites of the mutton sheep based on machine vision, which is characterized in that the self-adaptive region-of-interest extraction submodule adopts a light convolutional neural network as a region proposal network, the region proposal network consists of 3 convolutional layers and 2 fully-connected layers, the network input is a preprocessed whole sheep body surface image, the network input is output as coordinates of a plurality of candidate rectangular regions in the image and scores of the candidate rectangular regions serving as potential parasite perching regions, and the self-adaptive region-of-interest extraction submodule only keeps the first 5 candidate regions with the scores exceeding a preset threshold value in actual operation and takes the first 5 candidate regions as input of a subsequent detail enhancement flow.
6. The intelligent recognition system for the ectoparasites of the mutton sheep based on machine vision is characterized in that the implementation process of the multi-scale image fusion algorithm based on the Laplacian pyramid is that firstly, a Gaussian pyramid is built for an input candidate region image, then, the Laplacian pyramid of each layer of image of the Gaussian pyramid, namely, the difference value between the layer of image and the upper layer of image after being sampled is calculated, then, for a specific layer of the Laplacian pyramid, the high-frequency detail component of the Laplacian pyramid is enhanced by adopting an adaptive gain coefficient which is dynamically adjusted according to local gradient information of the layer of image, and finally, the enhanced Laplacian pyramid is combined with a bottom layer image of an original Gaussian pyramid, and a detail enhanced high-definition focusing sub-image is synthesized through a pyramid reconstruction process.
7. The intelligent recognition system of the meat sheep ectoparasites based on machine vision, which is characterized in that the hierarchical convolutional neural network model is trained by adopting an improved loss function, the improved loss function is formed by weighted summation of three parts of positioning loss, classification loss and focusing loss specially designed for tiny targets, the positioning loss adopts a smooth L1 loss function, the classification loss adopts a cross entropy loss function, and the focusing loss is a dynamically weighted cross entropy loss, and automatically reduces the weight of a sample easy to classify in the total loss through a modulation factor.
8. The intelligent recognition system for the ectoparasites of the mutton sheep based on machine vision according to claim 1, wherein the mapping table on which the rule engine is based is a multi-dimensional lookup table, an index key of the lookup table comprises a parasite type code and a life cycle stage code, a data structure stored in each table item comprises a risk grade field, a suggested chemical treatment scheme field, a suggested physical treatment scheme field and an early warning threshold field, the risk grade is divided into low, medium and high levels, and the early warning threshold field defines a lower limit of infection rate for triggering swarm early warning.

Description

Mutton sheep ectoparasite intelligent recognition system based on machine vision Technical Field The invention relates to the technical field of image recognition, in particular to an intelligent recognition system for ectoparasites of mutton sheep based on machine vision. Background The machine vision technology is taken as an important branch in the artificial intelligence field, is widely applied to multiple industries such as agriculture, industry, security protection and the like, realizes target detection, identification and classification through image acquisition and intelligent analysis, and remarkably improves the automation level of production and management, wherein an image identification method based on deep learning has become a mainstream technical direction for solving complex visual tasks due to strong feature extraction and pattern identification capability. The patent publication No. CN110032973A discloses an artificial intelligence based method and system for classifying an unsupervised parasite, which comprises the steps of obtaining a training data set of a sample to be detected, extracting characteristic information of the training data set by utilizing a deep convolutional neural network VGG network, wherein the VGG network is a standard VGG network or a pre-trained VGG network, classifying the characteristic information by utilizing a fuzzy C-means clustering FCM algorithm to determine a clustering center matrix of each category, and determining a clustering center vector of each category according to the clustering center matrix. As shown by the technology, in the field of animal husbandry health management, particularly for detecting ectoparasites of mutton sheep, the traditional method mainly relies on manual visual inspection or laboratory microscopic inspection, the manual inspection is low in efficiency and high in subjectivity and is easy to miss, while the laboratory method is accurate but long in flow and time consumption, the requirements of a large-scale farm on rapid epidemic screening and real-time monitoring cannot be met, although research is attempted to apply a general target detection model to animal ectoparasite identification, the prior art scheme still has the remarkable challenges that the mutton sheep breeding environment is complex, image background interference is serious, parasite target size is tiny and changeable, so that the detection precision and robustness of the general model are insufficient, in addition, parasite appearance characteristic differences in different growth stages are remarkable, the single model is difficult to realize effective identification of a whole life cycle, the model generalization capability is limited, the real-application scene brings higher requirements on the real-time performance and deployment convenience of an identification system, the existing high-precision model is large in quantity and complex in calculation, the conventional high-precision model is difficult to operate efficiently on edge computing equipment, and the lightweight model is often at the expense of sacrificing precision, and meanwhile, the focus system is difficult to accurately distinguish the defect from the defect of the surface of the parasitic object and the defect in the accuracy and the defect of the defect detection. Disclosure of Invention Aiming at the defects of the prior art, the invention provides an intelligent recognition system for the ectoparasites of the mutton sheep based on machine vision, which solves the problems that the existing intelligent recognition system for the ectoparasites of the mutton sheep lacks a focusing amplification and detail enhancement mechanism for suspected targets, is difficult to accurately distinguish parasites from similar foreign matters on the surface of complex hair, and has high false detection and omission rate. The intelligent recognition system for the ectoparasites of the mutton sheep based on machine vision comprises an image acquisition module, a recognition module and a recognition module, wherein the image acquisition module is used for acquiring original visual data of the body surface of the mutton sheep; The image preprocessing and enhancing module is connected to the image acquisition module and is used for carrying out standardization processing on an original image and carrying out region enhancement facing to a tiny target, the image preprocessing and enhancing module comprises an adaptive region of interest extraction submodule and a detail enhancing flow, the adaptive region of interest extraction submodule is used for initially positioning regions with dense hair textures or abnormal color plaques in the image based on a pre-trained shallow neural network and judging the regions as potential parasite habitat areas, and the detail enhancing flow is used for processing each extracted potential habitat image by adopting a multi-scale image fusion algorithm based on a Laplace pyramid