A novel pig-body multi-feature representation method based on multi-source image fusion

The detection of pig-body health has been a hot topic since it is closely related to the safety of agricultural products quality in recent years. In this paper, shape and temperature feature are viewed as health representation, and the multi-source image fusion method should be used to obtain pig-body multi-feature. However, traditional infrared and visible fusion algorithm could not represent more thermal radiation and texture information of fused images. To more usefully represent pig-body mulit-feature, a novel multi-source image fusion method is proposed in view of Generative Adversarial Network (GAN), named as MGANFuse. Firstly, multi-source images are fused by modified GAN fusion framework. Then, based on fused images, pig-body shape feature is represented by automatic threshold of Otsu segmentation and morphology method. Finally, maximum temperature feature is obtained based on pig-body shape representation. Experimental results reveal that representation method in view of proposed fusion model is capable of realizing 1.551–3.876% higher average accuracy rate than more recently published algorithms. Moreover, it lays the foundation for effective representation of pig-body multi-feature.