عنوان مقاله [English]
Introduction: Sistani cows are generally restless animals; therefore, controlling, treating, and weighing them is difficult. On the other hand, recording the weight of domestic animals, including Sistani cows, is inevitable, because it provides a good scale for management decisions in the herd such as balancing the diet, changing environmental conditions, or determining the time of slaughter of fattening animals. In addition to scales, various methods are commonly used to measure the body weight of large animals. Some of these methods include the use of weight-meters, appraisal assessments, and the use of mathematical models. One of the new methods for predicting livestock weight is artificial intelligence. Because some reports are indicating that artificial intelligence could facilitate the weighing process of animals, this study was performed to predict the body weight of Sistani cows using computer vision technology.
Materials and methods: The data required for this study were recorded in the Zahak breeding station located in Sistan and Baluchestan province of Iran. The recording operation involved the weighing and biometric measurement of about 190 Sistani cattle, including calves, heifers, and male and female animals, every three months during a year. At the time of weighing, images of the lateral view of each animal were taken and recorded using the CANON SX150IS digital camera. During this period, a total of 358 weight records of Sistani cows at different ages were recorded. The digital images were initially preprocessed using MATLAB software, and then some morphological features were extracted from each image. For predicting the weight of Sistani cows via the Artificial Neural Network (ANN), the extracted features of images were introduced to the ANN model as input and the weight of cows as output. The "feed-forward neural network", which was trained by the "error propagation" algorithm, was used to predict the weight of cows. The function used in the hidden layer of the ANN model was sigmoidal and in the output layer was linear. An ANN model which had the highest precision and lowest error was selected as the final model for predicting the animal weights. The criteria for selecting the best model were the highest determination coefficient (R2) and the lowest mean square error (MSE) compared to other available models.
Results and discussion: Out of 22 features extracted from each image, only 15 of them, which had a higher correlation with the body weight of cows at different ages, were selected as effective features. As result, equivalent diameter, major axis length, minor axis length, bounding box, convex area, filled area, area, perimeter, and the number of non-zero pixels of the image (NNZ) had the highest correlation with the cattle weights (P<0.01) and used as effective features to train the ANN model. The final ANN model had 15 neurons in the input layer including selected image features, 11 neurons in the hidden layer, and one neuron in the output layer including the weight of the cows. The precisions of the artificial neural network in the training, validation, and test phase were 0.974, 0.970, and 0.981, respectively. The results showed that the final ANN model had acceptable precision in all light, medium, and heavy-weight cows, and the size and the age of animals did not have a significant effect on the precision of the artificial neural network model. A correlation between the actual weight of Sistani cows and the weights predicted by the ANN model was 98.3%. The average error of the model in predicting the weight of cows was 1.11%. In the practical test, a 2.32 kg deviation was observed between the predictions of the ANN model and actual weights in Sistani cows. The accuracy of the ANN model for predicting the weight of Sistani cows in the present study is acceptable and within the range of the other reports.
Conclusions: The proposed method based on image processing and ANN, had acceptable results in predicting the weight of Sistani cows. Given the difficulties of weighing Sistani cows as heavy livestock and sometimes the time-consuming process, it seems that the use of new technologies such as computer vision methods can be a good alternative to conventional weighing methods and facilitate and reduce recording costs of Sistani cows.