Animal image target detection system for livestock industry in Qinghai plateau

Abstract

This invention proposes a target detection system for animal images in the Qinghai Plateau livestock industry, comprising: an opening module, a target detection module, and a target statistics module. The target detection module is based on the YOLOv3 model. It first introduces the k-means++ clustering algorithm to re-cluster the dataset and select ideal anchor values, then adds a (104*104) feature scale, subsequently improves the Non-Maximum Suppression (NMS) algorithm, and finally introduces KL-Loss to calculate the bounding box regression loss. This invention achieves animal image recognition and target detection in the livestock industry based on a deep learning network model, significantly improving detection accuracy.

Description

Technical Field

[0001] This invention relates to the field of target detection technology, and in particular to a target detection system for animal images in the Qinghai Plateau livestock industry. Background Technology

[0002] Animal husbandry is a major economic source in Qinghai Province, and its development directly impacts the living standards of farmers and herders. Currently, Qinghai Province employs various animal husbandry management models, including individual household grazing, joint household grazing, and hired grazing. Manual management of livestock not only wastes time and human resources but also suffers from omissions in statistical and identification processes. Therefore, a livestock target detection system, using mechanized methods to monitor livestock activity, can effectively monitor livestock conditions. This not only reduces human and material resources but also minimizes safety risks for managers and inspectors, making animal husbandry management more modern and information-based in both methods and standards.

[0003] Qinghai's unique climate results in significant seasonal variations in its natural environment, leading to differences in the coat color and body shape of livestock animals. Unlike other regions or livestock breeding areas, existing image target detection systems struggle to classify and identify these animals effectively, resulting in low accuracy. Therefore, improvements are urgently needed. Summary of the Invention

[0004] To address the shortcomings of the existing technologies, this invention proposes an improved F-YOLOV3-based target detection algorithm for animal images from the Qinghai Plateau.

[0005] The technical solution of this invention is implemented as follows:

[0006] A target detection system for animal images in the Qinghai Plateau livestock industry includes: an opening module, a target detection module, and a target statistics module. The opening module reads the image into the system, and then the target detection module detects the image and selectively outputs the detected targets. If it is necessary to perform statistics on the detected targets, the target statistics module performs statistical operations on the detected targets and selectively outputs the statistical results.

[0007] The target detection module is based on the YOLOv3 model:

[0008] First, the k-means++ clustering algorithm is introduced to re-cluster the dataset and select ideal anchor values ​​to improve the predicted bounding boxes;

[0009] Furthermore, based on the three existing feature sizes of the YOLOv3 model: (13*13), (26*26), and (52*52), an additional feature size of (104*104) is added;

[0010] The non-maximum suppression algorithm (NMS) was then improved.

[0011] Define f(x) = 1 - tanh(x) 2 / σ);

[0012] Among them, s i N represents the final score of the predicted bounding box, and IOU is the intersection-union ratio between the predicted bounding box and the detection box with the highest confidence. t For setting a predefined threshold;

[0013] Finally, KL-Loss is introduced to calculate the bounding box regression loss, and the calculation formula is as follows:

[0014]

[0015]

[0016] Preferably, the target detection module includes the following sub-modules: input image, preprocessing, predicted bounding box, predicted category, and output result. The preprocessing performs size and format conversion on the image to be detected. The predicted bounding box is determined based on the coordinate position and confidence information of the image to be detected. The predicted category is determined by a classifier.

[0017] Preferably, the target statistics module includes the following sub-modules: input image, target recognition, candidate box generation, and counting of the number of targets in each category and output results, wherein the counting of the number of targets in each category is performed according to the predicted category and probability value.

[0018] This invention realizes livestock animal image recognition based on a deep learning network model and realizes livestock animal image target detection based on an improved deep learning network model, which significantly improves the detection accuracy. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the process of the present invention;

[0021] Figure 2 This is a block diagram of the target detection module of the present invention;

[0022] Figure 3 This is a block diagram of the target statistics module structure of the present invention;

[0023] Figure 4 for Figure 2 The target detection module's detection results are shown in the image.

[0024] Figure 5 This is a diagram showing the detection results from the original target detection module;

[0025] Figure 6 for Figure 3 The statistical results of the target statistics module are shown in the figure.

[0026] Figure 7 This is a graph showing the statistical results of the original target statistics module;

[0027] Figure 8 The test results of this invention Figure 1 (Multi-seasonal);

[0028] Figure 9 The test results of this invention Figure 2 (Multi-objective) Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] like Figure 1 As shown: The first step in using this system is to prepare data by saving the selected photos to an easily accessible location. If images from the test set are used, preprocessing is unnecessary. However, if new images are used as target images, preprocessing is required. After preparation, the images can be tested on the system. The images are read into the Qinghai Plateau Livestock Animal Image Target Detection System by opening the function module and displayed in the original image display area. The target detection module then performs detection operations on the read images. After a series of operations within the system, the detected targets are output in the image results display area. If no statistical analysis is needed, the operation can be terminated. However, if statistical analysis of the detected targets is desired, the statistical function module performs statistical analysis on the detected targets, outputting the results according to category and total count.

[0031] In this animal image target detection system for animal husbandry on the Qinghai Plateau, the target detection module is the core component of the entire system. The target detection module is based on the YOLOv3 model: it first introduces the k-means++ clustering algorithm to re-cluster the dataset and select ideal anchor values, thereby improving the predicted bounding boxes.

[0032] Furthermore, based on the three existing feature sizes of the YOLOv3 model: (13*13), (26*26), and (52*52), a feature scale of (104*104) is added; subsequently, the Non-Maximum Suppression (NMS) algorithm is improved.

[0033] Define f(x) = 1 - tanh(x) 2 / σ); where s i N represents the final score of the predicted bounding box, and IOU is the intersection-union ratio between the predicted bounding box and the detection box with the highest confidence. t A predefined threshold is set; finally, KL-Loss is introduced to calculate the bounding box regression loss, and the calculation formula is:

[0034]

[0035] The improved target detection model for animal images on the Qinghai Plateau, obtained through the above methods, can effectively improve the detection accuracy of small targets.

[0036] For images to be input into the system for detection, preprocessing is required first. This is because the system pre-sets image size and format; if an unprocessed image with a different format and size is input, the system will be unable to recognize it, resulting in detection failure. The preprocessed image, which the system can correctly recognize, undergoes a series of algorithmic operations to determine the bounding box of the target based on its location coordinates and confidence level. Then, a classifier predicts the category of the determined bounding box. Finally, the predicted category information and probability value of the detected target are output in the image results display area. The target detection module structure diagram is shown below. Figure 2 As shown.

[0037] The initial purpose of designing the statistical function module in this Qinghai Plateau livestock animal image target detection system was to save time. During target detection in images, it was found that some livestock animal images exhibited overlapping and interleaved targets due to excessive density. Manual target counting is not only inefficient but also time-consuming and labor-intensive. Based on this problem, a target counting function module was designed to make target counting more convenient and effective. The structure of the target counting function module is as follows: Figure 3As shown.

[0038] As for the interface partitioning of this invention, it can be divided into 6 functional areas, namely: opening image, target detection, target statistics, original image display area, detected image result display area and result display area.

[0039] To verify the detection effect of the present invention, a single image taken in a natural environment was used as the test object to test and compare the detection performance of the improved model.

[0040] like Figure 4 and Figure 5 As shown, the target detection module of the present invention can detect all targets in an image, while the original target detection module identifies small targets (horse) in the image as yak. ​​It can be seen that the target detection module of the present invention improves the target detection accuracy of the image.

[0041] like Figure 6 and Figure 7 As shown, the target statistics module of this invention counted a total of 56 targets, while the original target statistics module counted 51 targets. Clearly, the statistical results further verify that the present invention does indeed improve the target detection accuracy.

[0042] To further verify the authenticity and validity of the above results, this invention also selected corresponding data according to different seasons, categorized and tested them, and statistically verified the authenticity of the above results. The figure includes the target category and confidence information, and the test results are as follows: Figure 8 As shown: The figure contains test results for four seasons: spring, summer, autumn, and winter. The first row of images is an example of summer, and the first column shows images with both large and small targets, including small targets that are easily identifiable by the human eye. The second row shows test results against a winter background, and the second column shows detection results characterized by overlap; it can be seen from the image that targets with a high degree of overlap are detected. The third row shows target detection results against a summer background, and the third column shows detection results mainly for small targets, such as the first image in the third column, where the system detects targets that are difficult for the human eye to distinguish. The fourth row shows detection results against an autumn background. In addition, this invention also performs multi-target detection, such as... Figure 9 As shown in the last image, the most typical example is that all three types of targets are present, and their actions and appearances are very similar. However, all three types of targets are correctly detected, and their confidence levels are all 1.

[0043] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. Qinghai plateau animal husbandry animal image target detection system, characterized by: Include: The opening module, target detection module and target statistics module, the opening module reads the image into the Qinghai plateau animal husbandry animal image target detection system, then the target detection module detects the read image and selectively outputs the detected target, if the detected target needs to be counted, the target statistics module is used to count the detected target and selectively output the counting result; The target detection module is based on YOLOV3 model: Firstly, the k-means++ clustering algorithm is introduced to re-cluster the data set and select the ideal anchor value, so as to improve the prediction box; Secondly, (104*104) feature scale is added on the basis of the three existing feature sizes of YOLOV3 model: (13*13), (26*26), (52*52); Then, the non-maximum suppression algorithm NMS is improved, define f(x) = 1 - tanh(x / σ) 2 / σ); wherein s i represents the final score of the prediction box, IOU is the intersection over union between the prediction box and the detection box with the largest confidence, N t is a predefined threshold set. Finally, KL-Loss is introduced to calculate the bounding box regression loss, and the calculation formula is: 2.The Qinghai plateau animal husbandry animal image target detection system according to claim 1, characterized in that: The target detection module includes the following sub-modules: input image, preprocessing, prediction bounding box, prediction class and output result, the preprocessing converts the size and format of the image to be detected, the prediction bounding box determines the bounding box according to the coordinate position and confidence information of the image to be detected, and the prediction class is classified by the classifier. 3.The Qinghai plateau livestock animal image target detection system of claim 2, wherein: The target statistics module includes the following sub-modules: input image, target recognition, candidate box generation and target number statistics of each class, and output result, the target number statistics of each class is counted according to the predicted class and probability value.

Images