Recognizing and monitoring infectious sources of schistosomiasis by developing deep learning models with high-resolution remote sensing images

Abstract

Abstract Objective: This study aimed to develop deep learning models with high-resolution remote sensing images for recognizing and monitoring livestock bovine, which is an intermedia source of Schistosoma japonicuminfection, and to evaluate the effectiveness of developed models for real-world application. Methods: The dataset of livestock bovine’s spatial distribution was constructed based on high-resolution remote sensing images. The images were further divided into training and test datasets for model development. Two recognition models based on deep learning methods (ENVINet5 and Mask R-CNN; were developed with reference to the training datasets. The performance of the developed models was evaluated by the performance metrics of precision, recall, and F1-score. Results: A total of 50 typical image areas were selected, and 1,125 bovine objectives were labeled by the ENVINet5 model and 1,277 bovine objectives were labeled by the Mask R-CNN model. For the ENVINet5 model, a total 1,598 records of bovine distribution were recognized. The model precision and recall were 81.85% and 80.24%, respectively. The F1 score was 0.807. For the Mask R-CNN mode, 1,679 records of bovine objectives were identified. The model precision and recall were 87.32% and 85.16%, respectively. The F1 score was 0.865. When applying the developed models to real-world schistosomiasis-endemicregions, there were 63 bovine objectives in the original image, 53 records were extracted using the ENVINet5 model, and 57 records were extracted using the Mask R-CNN model. The successful recognition ratios were 84.1% and 90.5% for the respectively developed models. Conclusion: For the control of schistosomiasis, it is feasible to recognize and monitor livestock bovine by developing deep learning models with high-resolution remote sensing images. The ENVINet5 model can use a few training images to create training datasets with a reasonable accuracy of segmentation. The ENVINet5 model is very feasible for when bovine distribution is low in structure with few samples. The Mask R-CNN model needs to create labels along the target object and requires a long time for data training. The Mask R-CNN model is good in the framework design and runs highly efficiently. The livestock recognition models developed using deep learning methods with high-resolution remote sensing images very accurately recognize the spatial distribution of livestock, which could enable precise control of schistosomiasis.