Authors

Abstract

Individual pig identification technology is the precondition of precise breeding. Taking pig face as the study point, this article puts forward a pig face identification method based on improved AlexNet model and explores the influence of training batch size on the performance of the model. Spatial attention module (SAM) is introduced in AlexNet model to compare the performance of the AlexNet model and the improved model on the training set and the validation set. The study shows that the improved AlexNet model can achieve higher precision rate under different training batch sizes and has higher convergence rate and robustness, with an identification precision rate reaching 98.11%, and a recall rate and f1 value reaching 98.03% and 98.05%. When the training batch sizes are 16, 32, and 64 respectively, the test time of the model, which represents its operating efficiency, improves by 1.99%, 2.36% and 10.31%, respectively, showing better performance in pig face identification. The test results show that different batch sizes have a certain influence on the prediction results of the model, while no fixed relationship.

Abstract in Chinese

生猪个体识别技术是实现精准养殖的前提，本文以猪脸为研究点提出基于改进AlexNet模型的猪脸识别方法并探究训练批大小对模型性能影响，在AlexNet模型中引入空间注意力模块（SAM），比较AlexNet与其改进模型在训练集及验证集上的性能，研究表明，改进的AlexNet模型在不同训练批大小情况下均可取得较高准确率，具有更快的收敛速度与鲁棒性，识别准确率达到98.11%，召回率与f1值分别达到98.03%、98.05%，其在训练批大小分别为16、32、64情况下，表征其运行效率的模型测试时间分别提高了1.99%，2.36%、10.31%，可对生猪脸部进行更有效识别，试验结果表明，不同的批大小对模型预测结果有一定影响，不存在固定关系。