Deployment Of Machine Learning Model to Evaluate the Quality of Green Spacing

Green space is any vegetated infrastructure. Green space promotes mental and physical wellness, socialization, and environmental protection. The state of green space determines its quality. Based on machine learning research, trash, lack of upkeep, and dirtiness lower green space quality. These algorithms evaluate green areas without considering human perception. Domain-based approaches are laborious, tedious and difficult to implement across large areas. This study suggests applying transfer learning on pre-trained models to design, analyze, and implement a machine-learning technique for human-perception green space quality assessment. Precision, F1-score, accuracy, recollection, Cohen’s Kappa, and Regular ROC-AUC were high for the generated models. In addition, the models’ file sizes and inference times were analyzed to guarantee their use in real-world settings. Grad-CAM, which uses heat maps to assess a model’s learning performance, was also included in the study. ResNet50 was the top-performing model, with 99.99 per cent accuracy, 99.99 per cent precision, 99.99 per cent recall, 99.00 per cent F1-score,0.98 Cohen’s Kappa, and an Average ROC-AUC of 1.00. The ResNet50 model was the second-fastest predictor and had a reasonable file size. Grad-CAM imaginings reveal that ResNet50 can accurately pinpoint regions crucial to its learning. Finally, an interactive online application based on the ResNet50 model was released via the Stream lit cloud platform.