Development and Optimization of Al model for Feature identification/ Extraction from drone orthophotos
ABSTARCT : Background 1. The hon’ble prime minister launched the svamitva scheme on the national panchayati raj day, 24th april 2020 with a resolve to enable the economic progress of rural india by providing “record of rights” to every rural household owner.
The scheme aims to demarcate inhabited (abadi) land in rural areas through the latest surveying drone technology, continuous operating reference system (cors), and geographic information system (gis) technology. The scheme covers multifarious aspects viz.
Facilitating monetization of properties and enabling bank loan; reducing property-related disputes; comprehensive village-level planning 2. With the use latest drone technology and cors technology for the abadi land survey, the high resolution and accurate image base maps of 50 cm have facilitated creation of the most durable record of property holdings in these areas with no legacy revenue records.
Such accurate image-based maps provide a clear demarcation of land holdings in a very short frame of time compared to on ground physical measurement and mapping of the land parcels.
Description i. Develop an ai model capable of identifying key features in orthophotos with high precision: use of ai/ml techniques for extraction of the following features from svamitva drone imagery using a cloud-based solution: - a. Building footprint extraction (built-up area from the drone image and classified roof-top based on observation on the imagery as rcc, tiled, tin, and others.
These built up area can be used for various service such as solar energy calculation, property tax calculation, etc.) B. Road feature extraction c. Waterbodies extraction, etc ii. Achieve a target accuracy of 95% in feature identification. Iii. Optimize the model for efficient processing and deployment.
The broad scope of work includes:- i. Data preparation: utilize the svamitva scheme drone-labeled datasets for 10 villages to train and validate the ai model.
Model development: employ convolutional neural networks (cnns) or other suitable ai/ml models for image segmentation and feature identification. Model training: efficient training of the model ensures it learns to identify the desired features accurately.
Deployment and integration: prepare the model for deployment and integration into the ministry of panchayati raj’s existing systems for practical application. Expected solution A fully trained and optimized ai model for feature identification in orthophotos.
Documentation detailing the model architecture, training process, and deployment guidelines. Iii. A final report summarizing the project outcomes, including accuracy metrics and recommendations for future improvements.
Model optimization: fine-tune the model to achieve the target accuracy of 95%, optimizing for speed and resource utilization. Testing and validation: rigorously test the model on a separate dataset to ensure its accuracy and reliability.
Existing methods can be classified in different ways. One possibility is to divide them into two groups – automatic and semi-automatic.
Semi-automatic methods – as opposed to the automatic ones – require human intervention, especially when tuning algorithms and judging results. Because of the many influences that contribute to the quality of aerial imagery, we usually cannot fully rely on automatic methods.
This is expressed in a strong tendency to combine different methods and algorithms for solving various problems. As will be described in detail in the following text, in our approach we proceed in the same way. For a description of images we use a set of general features (we do not use any knowledge base of known objects for extraction).
These features are detected by a set of specific algorithms. We use a neural network for tuning these algorithms. The features are then detected automatically.
Image quality: the quality of drone orthophotos can vary due to factors such as weather conditions, altitude, and the resolution of the camera used. Poor image quality can hinder the performance of ai models.
Resource-intensive training: training ai models, especially deep learning algorithms, requires significant computational power and memory. This can necessitate expensive hardware or cloud computing resources.
Overfitting: ai models can become overly complex and tailored to the training data, leading to poor generalization when applied to new or unseen data.
Black box nature: many ai models, especially deep learning models, operate as "black boxes," making it challenging to understand how decisions are made. This lack of transparency can hinder trust in the model’s outputs.
Our aim is automatically to identify similar image regions, which can be later semi-automatically explored. We start with an orthophoto image. This image has to be normalized to allow comparison with different orthophotos. Once normalized, we split the image into particular regions.
The size of the regions in calculated using image resolution and sensing distance and expected size of particular objects. In certain situations we include also an overlap of regions to minimize the previously mentioned problem of splitting ambiguity. Particular regions are then automatically processed using the described image classifier.
This results in a dataset of particular regions and identified image features. Now we can verify the validity of detected image features using supervised training and our experimental application. This task can also be used to select a suitable set of image features for further exploration. This step is required because the visual similarity is very subjective and context-dependent. We need to capture the user’s point of view.
Automated analysis: ai models can automate the process of identifying and extracting features, significantly reducing the time and effort required compared to manual methods.
Handling large datasets: ai models can efficiently process and analyze vast amounts of data generated by drone imagery, making it feasible to scale up operations without a significant increase in resource investment.
Handling large datasets: ai models can efficiently process and analyze vast amounts of data generated by drone imagery, making it feasible to scale up operations without a significant increase in resource investment.
Reduced labor costs: automating feature extraction reduces the need for extensive human labor in data processing and analysis, leading to cost savings.
|