Unveiling the green guardians: Mapping and identification of Azadirachta indica trees with semantic segmentation deep learning neural network technique

Abstract

The high spatial resolution data presents a problem when it comes to mapping and identifying distinct tree species based on the characteristics of their canopies. The deep learning Semantic Segmentation approach based on U-Network (U-Net.) artificial intelligence model that we provide here can recognize, and map Azadirachta indica trees canopy cover. This method trains its model by making use of image chips and labels of the item being segmented. The new testing images processed for multiple stages of pixel level of convolution and pooling operations. The sampling methods allow increase to make complete to make the recognized object on the image. The model’s ability to identify items based on canopy shape, structure, and pixel data makes it very useful for mapping and recognizing a single tree species as well as several tree species. The model validation results indicated an accuracy of 84–89 percent, which is regarded to be rather good. Based on ground census data, the overall accuracy of identification is 89 percent, F1 score 0.91–0.94, while the complete tree canopy validation (Intersection to Union) for canopy matching area is 0.79–0.89. The method has the potential to be utilised for identification, mapping of tree canopy. The approach has the potential to be used for important research initiatives i.e tree censuses and the identification and mapping of crop plant identification. The deep learning model used as inferences for automatization of the identification of the tree species helps to resolve identification and mapping based complex problems in agro-forestry allied fields.

Keywords

• U-Network
• Conventional Neural Network
• Canopy cover
• Convolution-Polling
• Data Augmentation