A farm management method based on a neural network

Abstract

The invention discloses a farm management method based on a neural network, and the method comprises the steps: collecting various images of a farm, and processing the images to form an image data set; Forming a target detector by adopting a YOLO network, wherein the target detector adopts three scales; Inputting the image data set into a YOLO network to complete training; enabling The trained target detector to be used for detecting and counting targets bred in a breeding farm, and distinguishing other types of species; And positioning the detected human head, controlling a camera to automatically focus the human head to extract a clear head portrait, and storing and recording the clear head portrait. According to the method provided by the invention, the types of outbound harmful speciescan be accurately judged, and farmers can conveniently judge whether to take defense measures or not; The method provided by the invention has the advantages that experimental verification proves that the accuracy is higher when small targets such as chickens are detected; People entering and exiting the breeding pen are positioned, the camera is controlled to automatically focus, clear images are extracted and recorded and stored, and theft prevention, damage prevention and poisoning prevention are effectively achieved.

Description

Technical field [0001] The invention belongs to the field of artificial intelligence, and specifically relates to a method for managing a farm based on a neural network. Background technique [0002] Due to the rise of artificial intelligence, intelligent management models have also been applied to the breeding field, mainly including the supervision of the number of pigs, chickens, ducks, cattle, and sheep raised in outdoor enclosures to avoid loss and economic losses. [0003] At present, many experts use computer vision technology, use traditional pattern recognition methods to identify targets, and then count the targets one by one. The core of traditional pattern recognition methods is image processing technology. One of the main advantages of image processing technology is that almost all targets can be identified. Although many scholars have proposed many algorithms for target recognition and counting, including frame difference method, optical flow method, morphological an...

AI Technical Summary

Problems solved by technology

Although many scholars have proposed many algorithms for target recognition and counting, including frame difference method, optical flow method, morphological analysis, mixed Gaussian background modeling, threshold segmentation, matching counting, etc., there are still many such recognition and counting methods. The problem is that the images under various actual shooting conditions contain various problems such as light intensity, light and dark changes, and image distortion. During the processing, a lot of noise will be generated, and the detection effect is more likely to be affected by the environment; in addition, The categories of traditional pattern recognition and classification are very limited, and are greatly affected by the shape of the target and the quality of the color image. Therefore, the recognition and classification have certain limitations, and so far there is no optimal solution

[0004] In response to the above problems, scholars have also made some improvements based on image processing feature extraction technology combined with machine learning methods, but there are still large errors

Images