We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Remote sensing image classification using modified random forest with empirical loss function through crowd-sourced data.
- Authors
Pazhanikumar, K.; KuzhalVoiMozhi, S. Nithya
- Abstract
Environmental changes are captured as satellite images and stored in datasets for monitoring a particular location. These remote sensing images can be employed in predicting valuable data for both urban planning and in land-use management. Moreover, infrastructure planning, management of natural resources are also assessed using these images. Many traditional classification approaches have been utilized in classifying images. However, it resulted with processing of limited images and acquired minimum accuracy rate. In order to overcome the challenges in existing systems, the present research takes effort in utilizing three datasets such as SAT-4, SAT-6 and RSI-CB for proposed classification. Using such crowd sourced data, images are annotated clearly through vector information. Such images are taken for three stage implementation in proposed work for effective classification of satellite images. Large scale distribution data contains six categories and 35 sub-classes with 40,000 images of size 128 × 128 pixels. Initially, pre-processing performs checking of missing values and feature scaling to avoid incorrect predictions. The second stage is the feature extraction process performed by VGG19 and ResNet 50 architecture which provides best features. Individually extracted features from both models are concatenated and sent as input to the third stage of classification. Modified RF (Random Forest) with empirical loss function is used for classification which classifies images with majority voting process on tree based structure. Loss values are computed to determine the efficacy of the model. In addition to that, classification process is also executed by DT (Decision Tree) and K-NN (K-Nearest Neighbour) to exhibit the classification efficiency of proposed model. Performance evaluation on three datasets along with comparative analysis with other classification methods in terms of precision, accuracy, f1-score and recall are performed for determining the effectiveness of present research.
- Subjects
IMAGE recognition (Computer vision); RANDOM forest algorithms; NATURAL resources management; FEATURE extraction; CROWDSOURCING
- Publication
Multimedia Tools & Applications, 2024, Vol 83, Issue 18, p53899
- ISSN
1380-7501
- Publication type
Article
- DOI
10.1007/s11042-023-17556-3