Description:[110] In this patent, plastic density detection in rivers - a generalized AI-based approach is proposed to reduce the plastic clearance cost and time from rivers and waterways and help in planning the process most efficiently. This innovative generalized AI system is developed utilizing image classification techniques to enable plastic pollution detection across various river ecosystems. This system is trained once and can be applied to various rivers from different geographies without the need for retraining. This innovation shows the plastic density percentage in the given image and highlights the plastic area. In addition to plastic, it shows the area and percentage of water, sand, and vegetation in the provided image. This invention raises a red, orange, and yellow alarm based on a dynamically pre-defined high-low threshold for plastic percentage. In this development, a plastic density exceeding 20% is categorized as high, and a red flag is raised, while less than 5% but more than 0% is considered low, and a yellow flag is triggered. An orange flag is displayed to inform the authorities to monitor and take appropriate action as necessary when there is more than 5% but less than 20% plastic density. To train the model, roughly 500–600 photos are gathered for each category: plastic, sand, water, and vegetation. Images are in 100 * 100 pixels JPG RGB colour code format. A comparable dataset consisting of 400 and 15 photos is designed for testing and validation, respectively. CNN architecture with five layers is used to train the model. The test dataset is used to test the model, while the validation dataset is used to validate it. The trained model is stored and available for use in making inferences. Three methods are applied and evaluated for inference. The image that is provided for inference is divided into blocks of 100 *100 pixels, 50*50 pixels, and 25*25 pixels. All the 100*100 pixels blocks are processed through the trained model to predict the class for each block. Similarly, all the 50*50 pixels blocks are given to the trained model to predict the class for each block, and the same approach is applied for all the 25*25 pixels blocks. Predicted classes for each 100*100 pixels block are displayed using a heatmap. The boxes for each 100*100 pixels tile are drawn using a heatmap, with different colours used for each class, water in blue, sand in yellow, vegetation in green, and plastic in grey. The same method is applied for 50*50 pixels blocks and 25*25 pixels blocks to visualize the water, sand, vegetation and plastic distribution and plastic density. A few further functions and rules are built to display the alert based on the low-high threshold set for plastic percentage. The provided image's longitude and latitude are also displayed to assist in locating the plastic portions that have been marked by this system using heatmap and visuals for the green plastic area's to be distinguished from the green vegetation using the predetermined colour channels. HSV (Hue, Saturation, and Value) format is also shown for the given image to help in planning the cleanup process.
This model requires fewer computing resources. In order to analyse generalization, this novel approach is evaluated using six river photos that were gathered from publicly accessible data sources for six distinct countries. The results show promise in terms of generalizability and applicability. It achieved more than 60% accuracy in predicting the water, sand, vegetation, and plastic in all the supplied images from 6 rivers from 6 nations. This invention is capable of adapting to diverse river ecosystems. By utilizing AI technology, it hopes to contribute to the worldwide effort to create waterways free of plastic.

, Claims:1. Implementation of “Plastic Density Detection in Rivers - Generalized AI-Based Approach” comprising:
a. A generalized system to detect plastic density in any river having similar nature and colour of debris.
b. Locate areas of plants, water, and sand in addition to plastic in rivers and waterways.
c. Plastic density detection system utilizing convolution neural network-based image classification approach.
d. Indicates a red (if plastic density in the given image is exceeding 20%), orange (if plastic density in the given image is more than 5% but less than 20%), or yellow (if plastic density in the given image is less than 5% but more than 0%) flag depending on the estimated percentage of plastic density in the given image for low-high thresholds defined for location-specific criteria.
e. Display longitude and latitude information of the image, if the longitude and latitude are captured while taking the picture, to help in locating the regions raised by the system.
f. Display green plastic objects to differentiate from green vegetation based on the RGB colouring ratio set to differentiate them.

2. The Implementation of “Plastic Density Detection in Rivers - Generalized AI-Based Approach” as claimed in 1 wherein a generalized system to detect plastic density in any river having similar nature and colour of debris using Machine Learning with Python codes.

3. The Implementation of “Plastic Density Detection in Rivers - Generalized AI-Based Approach” as claimed in 1 wherein locate areas of plants, water, and sand in addition to plastic in rivers and waterways using Machine Learning with Python codes.

4. The Implementation of “Plastic Density Detection in Rivers - Generalized AI-Based Approach” as claimed in 1 wherein plastic density detection system utilizing convolution neural network based image classification approach using Machine Learning with python codes.

5. The Implementation of “Plastic Density Detection in Rivers - Generalized AI-Based Approach” as claimed in 1 wherein it indicates a red (if plastic density in the given image is exceeding 20%), orange (if plastic density in the given image is more than 5% but less than 20%), or yellow (if plastic density in the given image is less than 5% but more than 0%) flag depending on the estimated percentage of plastic density in the given image for low-high thresholds defined for location-specific criteria using python codes.

6. The Implementation of “Plastic Density Detection in Rivers - Generalized AI-Based Approach” as claimed in 1, wherein the display of longitude and latitude information of the image, if the longitude and latitude are captured while taking the picture, to help in locating the regions raised by the system using python codes.

7. The Implementation of “Plastic Density Detection in Rivers - Generalized AI-Based Approach” as claimed in 1, wherein display green plastic objects to differentiate from green vegetation based on the RGB colouring ratio set to differentiate them using Python codes.