基于多模态融合技术的高精度半开放式种猪性能测定站设计与试验

Design and Experiment of High-precision Semi-open Pig Performance-Testing Station Based on Multi-modal Fusion Technology

摘要:

为提升种猪性能测定的自动化水平并减少外界干扰对测定精度的影响，设计了一种基于多模态融合技术的高精度半开放式种猪性能测定站。采用模块化设计原理，完成了测定站结构设计，将结构分成采食端和体质量称量端两个模块；基于STM32主控芯片与FreeRTOS操作系统完成了控制系统软硬件的搭建，与图像采集设备协同工作；提出了一种基于多模态融合的称量异常检测算法，利用背景减除与颜色过滤技术检测区域内异常，同时结合称量信号方差与极差进行动态分析优化时间窗口以对称量噪声数据进行滤除；开展了实际猪群养殖试验，结果显示，称量数据经去噪后，1号猪舍78.39%的异常数据和2号猪舍76.68%的异常数据恢复为正常数据，在30~60kg、30~100kg阶段，试验猪群日均自由采食次数分别为5.30（1号猪舍）、5.29（2号猪舍）和5.92（1号猪舍）、5.90（2号猪舍），日均采食时间分别为58.43min（1号猪舍）、63.23min（2号猪舍）和52.01min（1号猪舍）、54.95min（2号猪舍），料肉比分别为2.58（1号猪舍）、2.57（2号猪舍）和2.82（1号猪舍）、2.75（2号猪舍），80~140日龄的生长曲线拟合结果符合Logistic模型，符合猪群生长规律。研究结果表明，该测定站能够可靠进行种猪性能测定，为精准育种与现代化养殖提供了有效支持，对提升种猪选育效率和优化种猪遗传改良具有重要的应用价值和推广意义。

Abstract:

In order to improve the automation level of pig performance measurement and reduce the influence of external interference on the measurement accuracy, a high-precision semi-open pig performance testing station based on multi-modal fusion technology was designed. Employing the modular design principle, the structure of the testing station was devised. It was partitioned into two modules: the feeding end and the body-weight weighing end. Leveraging the STM32 main control chip and the FreeRTOS operating system, the hardware and software of the control system were constructed to operate in tandem with the image acquisition equipment. A weighing anomaly detection algorithm based on multi-modal fusion was proposed. This approach employed background subtraction and color filtering techniques to pinpoint anomalies in the designated area. Concurrently, it integrated the variance and range of weighing signals for dynamic analysis and optimization of the time window, which effectively filtered out the noise in the weighing data. The test of the actual pig herd was carried out, and the results showed that after the weighing data was denoised, 78.39% of the abnormal data of No.1 pig house and 76.68% of the abnormal data of No.2 pig house were restored to normal data. The results showed that in the 30~60kg and 30~100kg stages, the average daily ad libitum feeding times of the experimental herd were 5.30 (No.1 pig house), 5.29 (No.2 pig house), 5.92 (No.1 pig house) and 5.90 (No.2 pig house), respectively. The average daily feeding time was 58.43min (No.1 pig house), 63.23min (No.2 pig house), 52.01min (No.1 pig house) and 54.95min (No.2 pig house), respectively. As for the feed-to-meat ratios (FCR), it was 2.58 (No.1 pig house) and 2.57 (No.2 pig house) for 30~60kg stage, 2.82 (No.1 pig house) and 2.75 (No.2 pig house) for 30~100kg stage. Moreover, the fitting growth curves at the age of 80~140d were in line with the Logistic model, which was consistent with the growth patterns of the herd. The research findings indicated that this testing station was capable of reliably evaluating the performance of breeding pigs. It offered substantial support for precision breeding and modern swine production practices, which had important application value and promotion significance for improving the efficiency of breeding pigs selection and optimizing the genetic improvement of breeding pigs.