CN-122023914-A - Construction method and detection method of chicken follicle ultrasonic image automatic detection model based on deep learning

Abstract

The invention relates to the technical field of biological breeding and artificial intelligent image recognition, in particular to a method and a system for constructing an automatic detection model of chicken follicular ultrasonic images based on deep learning. Aiming at the problems that when a conventional target detection model is adopted to process an ultrasonic image of a chicken follicle under the condition that the specific structural optimization is not carried out in the prior art, the boundary identification of the follicle is unclear, the small-size follicle is easy to miss, the adjacent follicle is easy to identify as a single target in the ultrasonic image, and the like, the model is improved, the improved model can more accurately identify the central echo region of the follicle and the edge structure of the central echo region of the follicle, the adjacent follicle can be respectively marked, and the locating result of the follicle in the ultrasonic image is clearer and more stable.

Inventors

• HU XIAOXIANG
• GE MINGLIANG
• WANG YAOJUN
• LI XIANGDONG
• WANG ZHENGKUN
• HU ZEKUN

Assignees

• 中国农业大学

Dates

Publication Date

20260512

Application Date

20260202

Claims (9)

1. 1. The method for constructing the chicken follicle ultrasonic image automatic detection model based on deep learning is characterized by comprising the following steps of: s1, constructing a data set, namely constructing and marking a chicken follicle ultrasonic graphic data set based on a chicken abdominal cavity ovary region for training, verifying and testing a model; S2, feature extraction and model training are carried out by adopting an improved YOLOv model to carry out feature extraction and training to obtain an automatic detection model of chicken follicular ultrasonic images based on deep learning, wherein the improvement comprises the steps of introducing a downsampling module based on Haar wavelet transformation into a downsampling structure of a YOLOv model backbone network, introducing a space self-adaptive feature modulation module in a multi-scale feature fusion process of a neck network, and adopting a frame-free target detection mode in a detection head.
2. 2. The method for constructing a model according to claim 1, wherein the constructing and labeling of the chicken follicular ultrasound graphic dataset based on the chicken abdominal ovary region in S1 comprises four steps of ultrasound image acquisition, image processing, image labeling, and dataset partitioning.
3. 3. The method for constructing a model according to claim 2, wherein the image processing is to extract and screen the acquired ultrasonic image by using a time layered sampling method.
4. 4. The method for constructing a model according to claim 2, wherein the image labeling is to label the processed images one by using LabelImg labeling tool.
5. 5. The method according to claim 1, wherein the features extracted in the step of extracting the features S3 are a central hyperecho feature, an edge weak echo feature, and a spatial distribution feature of the echo intensity changing from the center to the edge of the chicken follicle dataset, and the features are extracted by the improved downsampling module in the step S2.
6. 6. The model construction method according to claim 1, wherein the training process in S3 is that the constructed and labeled chicken follicular ultrasound image dataset is input into an improved YOLOv model for supervision training, labeled follicular target position information and class information are used as real labels, model parameters are optimized through classification loss and bounding box regression loss, and finally the automatic detection model of the chicken follicular ultrasound image based on deep learning is obtained.
7. 7. A deep learning-based automatic detection model for chicken follicular ultrasound image obtained by the method for constructing a deep learning-based automatic detection model for chicken follicular ultrasound image according to claims 1 to 6, comprising the following modules: the data input module is used for inputting ultrasonic graphic data of chicken follicles in the ovary area of the abdominal cavity of the chicken; The data processing module is used for extracting and processing the characteristics of the input data; And the data output module is used for outputting the processed image to obtain the identification result of whether the image to be detected is a chicken follicle.
8. 8. An automatic detection method for chicken follicle ultrasonic images is characterized in that the model of claim 6 is adopted, a B ultrasonic image of only the abdominal cavity region of a target chicken to be detected is input, and the identification result of whether the input image is chicken follicles or whether chicken follicles exist is obtained after model processing.
9. 9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor for executing the model of claim 7.

Description

Construction method and detection method of chicken follicle ultrasonic image automatic detection model based on deep learning Technical Field The invention relates to the technical field of biological breeding and artificial intelligent image recognition, in particular to a method and a system for constructing an automatic detection model of chicken follicular ultrasonic images based on deep learning. Background Chickens are important livestock and poultry varieties, and the egg laying performance of the chickens is directly related to the genetic breeding efficiency and the industrial economic benefit. Egg yield is a core indicator for measuring egg production performance, and the number of follicles in ovaries and the hierarchical development state thereof are key biological factors for determining egg production potential. Therefore, the accurate phenotype information of the ovarian follicles of the individuals is obtained, and the method has important significance for improving the breeding accuracy in the poultry breeding process. However, in the current breeding scenario, it is still common to rely on slaughter measurement to obtain ovarian and follicular data, i.e. by dissecting hens to observe follicular morphology and develop sibling selection accordingly. The traditional method has the advantages of destructiveness, high cost and high labor intensity, continuous monitoring of the same breeding individuals cannot be realized, and the improvement of the breeding efficiency is greatly limited. In recent years, the development of ultrasound imaging technology has provided a non-invasive means for in vivo structural observations. Ultrasonic imaging can display tissue structures in vivo in real time, and is widely used for organ detection, focus positioning and auxiliary diagnosis in the medical field. With the maturation of image intelligent analysis technology, the target detection method based on deep learning has realized automatic identification and quantitative measurement in multiple medical ultrasonic images, and the detection precision and stability of the method are obviously improved. Although the combination of ultrasound and target detection techniques has found good application in the medical field, in-vivo nondestructive testing studies on chicken follicles are still very limited in the direction of poultry breeding. The existing work is focused on the identification of the tissue structure of mammals, and for the targets of small structure, dense distribution and large morphological difference of chicken follicles, a mature detection model and a complete set of technical flow are not yet available. Therefore, the construction of the automatic detection technology suitable for the chicken follicle ultrasonic image has important significance for breaking through the bottleneck of traditional slaughter measurement and realizing nondestructive accurate assessment of the ovarian development state. Disclosure of Invention Aiming at the problems that the ultrasonic image is mainly observed manually in the ultrasonic detection process of chicken follicles, the identification efficiency is low, the subjectivity is strong, the accurate positioning of the follicular structure in a complex ultrasonic background is difficult, and the like, the invention provides a method for returning the ultrasonic image of chicken follicles based on deep learning, which comprises the steps of constructing an automatic detection model of the ultrasonic image of chicken follicles based on deep learning, the method is used for detecting chicken follicles, can automatically identify and position the follicles in the ultrasonic image of chicken follicles on the premise of not damaging chicken, and clearly marks follicles from a complex ultrasonic background, so that visual and objective follicles identification results are provided for breeders, and the breeding personnel are assisted in carrying out follicular observation and relevant breeding analysis. In order to achieve the above purpose, the first technical scheme of the application discloses a method for constructing an automatic detection model of chicken follicular ultrasound images based on deep learning, which is characterized by comprising the following steps: s1, constructing a data set, namely constructing and marking a chicken follicle ultrasonic graphic data set based on a chicken abdominal cavity ovary region for training, verifying and testing a model; S2, feature extraction and model training are carried out by adopting an improved YOLOv model to carry out feature extraction and training to obtain an automatic detection model of chicken follicular ultrasonic images based on deep learning, wherein the improvement comprises the steps of introducing a downsampling module based on Haar wavelet transformation into a downsampling structure of a YOLOv model backbone network, introducing a space self-adaptive feature modulation module in a multi-scale feature