Mastitis Detection of Online Quarter-Milk Conductivity for Dairy Cows by Using the Artificial Neural Network

線上分房乳導電度類神經網路檢測泌乳牛乳房炎

連振昌(Cheng-Chang Lien) ； 萬一怒(Ye-Nu Wan) ； 顏名賢(Ming-Hsien Yen)

《農業機械學刊》 15卷1期 (2006/03) Pp. 1-13

https://doi.org/10.30062/JAM.200603.0001

摘要

使用一套乳房炎導電度線上檢測系統，能在擠乳過程中量測泌乳牛各分房乳導電度、分房乳間的導電度比值及乳溫的變化。實地試驗開始前及結束後採樣待測泌乳牛分房前乳量測體細胞數以作為健康分房及乳房炎分房的依據。由線上分房乳導電度及導電度比值應用一倒傳遞類神經網路來分類檢測泌乳牛的分房是否感染乳房炎，由乳房炎導電度線上檢測系統實地試驗所得到的資料，分組產生三個不同健康分房與乳房炎分房比例的訓練資料組，去訓練此類神經網路，並對四個不同健康分房與乳房炎分房比例的測試資料組進行驗證。由分析結果指出在訓練資料組中，較低的健康分房與乳房炎分房比例會使類神經網路有較佳的乳房炎分房預測能力，而較高的健康分房與乳房炎分房比例則會有較佳的健康分房預測能力；又在測試資料組中隨著健康分房與乳房炎分房比例的增加，真實陽性反應的預測機率P(PTP)顯著地遞減，但是其真實陰性反應的預測機率P(PTN)卻顯著地增加；此類神經網路能夠依據線上分房乳導電度及導電度比值分辨泌乳牛乳房炎，其正確反應的全部機率P(TCR)是88.2%。

關鍵字

乳房炎 ； 導電度 ； 體細胞數 ； 類神經網路

並列摘要

By using an online electrical conductivity (EC) measurement system for mastitis inspection of dairy cow, the variance of quarter-milk conductivity (QMC), ECR ratio among the quarter-milk conductivity and milk temperature can be measured during the milking. The somatic cell counts (SCC) of the foremilk from each quarter of dairy cows were measured at the day before and after of the field tests as the criterion to identify the healthiness of the udder. Therefore, it can be classified whether a quarter of dairy cow infected with mastitis or not according to the QMC and ECR indices which were used a back-propagation artificial neural network (ANN). All data of QMC and ECR were acquired from the field test by the online EC measurement system for mastitis would be classified three different ratios of healthy quarters to mastitis quarters (H/M ratio) in the training data sets, and be classified four different H/M quarter ratios in the testing data sets which would be executed to valid. The analysis results showed that lower H/M quarter ratio in the training data set had a better predictive probability of the ANN for mastitis quarters, and higher H/M ratio in the training data had a better predictive probability of the ANN for healthy quarters. In addition, as the H/M quarter ratio in the testing data increased, the predictive probability of true-positive response, P(PTP), decreased significantly, while the predictive probability of true-negative response, P(PTN), increased significantly. However, the probability of total correct response, P(TCR), of the ANN to identify the mastitis quarters was 88.2%. It is feasible to identify the mastitis cows according to online QMC and ECR of dairy cows using a back-propagation ANN.

並列關鍵字

Mastitis ； Electrical conductivity ； Somatic cell count ； Artificial neural network