Study And Implementation Of Cattle Body Detection Under Complex Background

Beef cattle breeding industry is of great significance in national economic construction, it is an industry with high-efficiency, section of grain and closely related to human livelihood, and is also the strategic direction of the adjustment of agricultural structure in China. As an important index to evaluate the quality of beef cattle, meat production is urgent needed to realize automatic measurement. Cattle detection as a preparatory work has great significance in meat production measurement. Aiming at the above problem, this paper analyzed the cow body images, and mainly adopted filter processing, cattle identification, morphological operation, and evaluation methods for image processing. Finally, the cattle body and complex natural background were separated successfully. This paper mainly includes the following sections:(1)Image prepossessing: Since in the process of cattle image acquisition, the images can be easily affected by light, then de-illumination and contrast enhancement are needed. In this paper, the Y-channel image was obtained from the cattle image in YCbCr space, and then homomorphic filter was used for filter processing, so as to enhance the image contrast. However, the removal for environment light impact is not obvious, and then two-level wavelet decomposition and homomorphic filter were adopted on in the Y channel image, which made the basic information be reflected in the lowest resolution layer, and the peak signal-to-noise ratio can reach 24.35. Therefore, using wavelet transform and homomorphic filtering methods under Y channel based on YCbCr space can reduce the effects of strong lights in the process of cattle body detection, and then enhanced the image contrast, which will provide preparation for the next cattle detection work.(2)Cattle body detection: This paper used two methods to extract the cattle body contour, which is bayesian classifier and improved Otsu method using skin detection and image segmentation approaches to separate cattle body from the complex background. In which we use bayesian classifier method in RGB and HSV color space to implementation of cattle body detection, Through statistics on different color channel detection effect, Establish the corresponding training sample and test sample.(3)Image optimization and evaluation: Complexity of cattle body environment and its own color leading to the background noise, then image optimization is needed. This paper uses morphological operation for image repairing and optimization, and then obtained the complete cattle body information. Furthermore, confusion matrix is used for evaluation and analysis of the optimization image. After analysis and calculation of 20 images, the experimental shows that the average accuracy under the bayesian method and the improved Otsu method are 86.17% and 80.07% respectively.This experiment solved the problem of separating the cattle body from the complex background, and achieved the cattle body contour extraction completely. As a result, it provided preparation work for subsequent cattle automation measurement.