Deep Learning with Improved Gravitational Search Optimization for Remote Sensing Image Classification

In earth observation technology using remote sensing (RS) data, RS image classification plays a significant role and has been extensively used in many domains. However, RS image classification presents substantial scientific and practical challenges because of the properties of RS data, such as high dimensionality and a need for more readily available labeled samples. Deep learning (DL) methods for RS image classification have advanced significantly in recent years as new DL approaches appear, opening up new opportunities for the study and growth of RS image classification for agricultural applications. The classification of RS images using an Improved Gravitational Search Optimization (IGSO) algorithm is discussed in this study. The field of organic carrots was collected in 60 top-down images by an autonomous vehicle and expertly labeled as part of the dataset for this study. Also provided is a comparison of the effectiveness of common DL-based RS approaches. Using 60 training images, IGSO outperforms existing methods by recognizing weeds with an accuracy of 98.6%. The challenges and prospective investigation paths are finally presented.