首页|基于自动摄像机与人工智能的奶牛跛行检测技术描述性评价

基于自动摄像机与人工智能的奶牛跛行检测技术描述性评价

扫码查看
[目的]通过评估基于自动摄像机(AUTO)的平均每周运动能力评分与奶牛首次出现病变的关系,探讨基于AUTO的运动能力评分是否可更早检测出奶牛首次出现跛行.[方法]自2022年4—12月,收集单个农场2982头奶牛的AUTO评分数据,包括奶牛ID、日期、时间和移动评分(0~100分).根据牧场历史记录中2204头奶牛的牛蹄病变数据,确定其病史和诊断日期.为消除慢性跛行的影响,研究重点关注无跛行史的奶牛,分为两类:首次确诊病变的奶牛(LESION)和修蹄师已检查但未确诊的奶牛(TRIM).确诊病变类别是根据修蹄时间进行诊断的,在干奶后7天内(DOT),或根据农场工作人员观察和修蹄建议随机(RT)进行.个体AUTO评分汇总为每周移动平均分.所有评分均以中位数(IQR)报告.按病变类型对LESION奶牛进行比较.[结果]DOT组(n=60)的病变类型:93%为TRIM、3.3%为趾溃疡(TOE),1.7%为白线病(WLD),1.7%为蹄底溃疡(SU).RT组(n=239)的病变类型:63%为TRIM,17%为蹄叶炎(DD),7.5%为蹄底溃疡(SU),7.1%为白线病(WLD),4.2%为腐蹄病(FR),4.2%为趾溃疡(TOE).RT组在前4周,LESION中位数评分(37.6[18.3])与TRIM(38.5[13.7])相似;在前1周,LESION中位数评分(41.1[17.5])高于TRIM(39.2[15.5]).DOT组在4周前,LESION中位数评分(59.2[2.1])高于TRIM(40.0[9.9]),这种模式一直持续到1周前.4周前,FR的评分最高(47.3[22.9]),随后分别是SU(42.8[19.0])、WLD(41.2[13.5])和DD(35.0[14.1]).1周前,FR(57.1[11.5])、SU(44.5[12.4])、WLD(44.3[26.8])和DD(39.5[10.6])的评分提高.[结论]AUTO评分或有助于早期识别某些病变.然而,不同奶牛个体和发病周数之间的差异性是一个有待解决的挑战.
Descriptive Evaluation of Camera-based Lameness Detection Technology Paired with Artificial Intelligence in Dairy Cattle
[Objective]The aim of this study was to explore whether autonomous camera-based(AUTO) mobility scores could detect first lameness occurrence earlier in cows,by assessing the association between average weekly AUTO mobility scores and cows with a lesion for the first time.[Method]The AUTO scores data were collected from 2982 cows in a single farm from April to December 2022,including cow ID,mobility score(0 to 100),and observation date and time.Historical farm hoof lesion data were collected from 2204 cows and used to determine cow lesion history and date of lesion diagnosis(LD).To remove the confounding impact of chronicity,the study focused on cows with no history of lameness and categorized them into two categories:those with a first-time LD (LESION) and those seen by a hoof trimmer without an LD(TRIM).These categories were compared based on when the trimming occurred:within seven days of dry off(DOT) or at a random time based on farm staff observation.Individual AUTO scores were summarized into moving average weekly scores.All weekly AUTO scores were reported as median[IQR].Comparisons were made for the LESION cows by lesion types.[Result]The lesion types for DOT(n=60) were 3.3% toe ulcer(TOE),1.7% white line disease(WLD),and 1.7% sole ulcer (SU),while the remaining had no reported lesion(93%;TRIM).For RT(n=239),63% were TRIM,17% digital dermatitis(DD),7.5% SU,7.1% WLD,4.2% foot rot(FR),and 4.2% TOE.Four weeks prior to RT,LESION had a similar median score(37.6[18.3]) to TRIM(38.5[13.7]).One week prior to RT,LESION had a higher median score(41.1[17.5]) compared to TRIM(39.2[15.5]).For DOT,four weeks prior,LESION had a higher median score(59.2[2.1]) than TRIM(40.0[9.9]),and this pattern persisted through 1 week prior.FR had the highest score(47.3[22.9]) four weeks earlier,followed by SU(42.8[19.0]),WLD(41.2[13.5]),and DD(35.0[14.1]).One week prior,these scores were increased for FR(57.1[11.5]),SU(44.5[12.4]),WLD(44.3[26.8]),and DD(39.5[10.6]).[Conclusion]The results suggest that AUTO scores may have the potential to detect some lesions earlier.However,there is variation between cows and weeks that presents a challenge yet to be addressed.

lameness detectionartificial intelligencedairy cattle

刘野、郭凯军

展开 >

北京农学院,北京 100096

跛行检测 人工智能 奶牛

2024

中国乳业
中国奶业协会 中国农业科学院信息研究所

中国乳业

影响因子:0.152
ISSN:1671-4393
年,卷(期):2024.(11)