AI to the rescue in the fight against Lantana Camara
Lantana Camara (popularly known as ‘balhari’ or ‘panchrangi’ in India) is an invasive species of plant that has grown and spread wide and deep within the forests of India. Lantana has its origin in South America and has spread wide and far across different parts of the world. Apparently, the plants have spread over 60 countries. Lantanas are flowering plants that are seen in most places today used mostly for ornamental purposes.
You must now be wondering if it is a beautiful flowering plant used for ornamental purposes then what is the problem with them growing or spreading in forest lands. Following is an excerpt from one of the recent articles on Lantana Camara and this should give us a peek into the problem we are dealing with.
Lantana not only spreads fast but does not allow grass, shrubs or any other plant in its vicinity to grow, leading to the migration or decline in the number of herbivores ultimately affecting the carnivores at the top of the food chain. The species is now able to climb up the canopy as a woody vine, entangle other plants by forming a dense thicket and spread on the forest floor as a scrambling shrub.
Invasive Species Specialist Group (ISSG) considers it as among 100 of the “World’s Worst” invaders. In India, Lantana Camara has been notorious for destroying entire local forest fauna, causing massive forest fires and blocking the natural reproductive cycle of forest trees. A recent study published in Global Ecology and Conservation reports that lantana occupies 154,000 sq.km forests (more than 40 percent by area) in India’s tiger range. Lantana management and control is the need of the hour. A combination of biological control, mechanical and chemical methodology could be used in Lantana management and control.
In every Tiger Reserve, a few hectares of land is cleared of Lantana Camara each year, but this cleared area requires intensive surveillance. Since lantana seeds are already present in the soil and since they are also dispersed by many birds from surrounding areas, Lantana regrows rapidly. To tackle this, a follow-up removal of Lantana Camara seedlings is necessary for a minimum of two years.
Spread of Lantana Camara in forests have also contributed to the forest fires in many forests around the world. In one of the recent forest fires that occured in the Bandipur forest reserve one of the officials has been quoted as saying — “If it wasn’t for lantana, the fire would have been 50%-60% less intense. It acts as fuel to fire. About 80% of the entire forest floor has been taken over by it.”
Fire-Lantana Cycle Hypothesis in Indian Forests by Ankila J. Hiremath and Bharath Sundaram in their study have - hypothesized that there may be a positive feedback between present-day fires and invasion by lantana, leading to a fire- lantana cycle that can have deleterious compositional and functional consequences for forest ecosystems and the commodities and services that society derives from them.
Based on existing literature and reports from the forest department there is a need for early detection and continuous monitoring of Lantanas in the forests.
Detection of Lantanas has been the key challenge to fight Lantanas. Through this article we illustrate a very simple use case for Lantana detection and reference some existing research work in this area where AI is used to detect Lantana Camara in forest lands, high lands, low lying areas, vegetations, agricultural lands, parks and other places.
This proposed simple use case aims at illustrating the use of AI to identify Lantana species using photographs/images taken by drones or otherwise, which is or can be used to monitor the forests for the Lantanas. A simple image recognition capability is suggested where an AI model (Tensorflow or Keras) is created and trained using a wide dataset of images of Lantana Camara and those that are not the L. Camara, but are other types of weeds. The dataset contains images of Lantanas in forest lands.
DeepWeeds project on GitHub is a great starting point to get such a dataset.
The model should be first trained. The trained models can then be used to predict Lantanas from images taken by monitoring equipments. The assumption is that the drones or other equipments used in realtime to capture the random images of the forests would also provide the geolocation of the images taken.
Proposed idea helps in the identification of Lantanas and reporting/alerting about their existence to the concerned forest officials. Once Lantanas are identified from realtime images, the forest officials can be alerted to de-weed the Lantanas at the location of its occurrence. Based on the frequency of monitoring and availability of realtime images the existence of Lantanas can be predicted and operations for their removal can be carried out before they grow to become a menace, thus, restoring the ecological balance in our forest lands and tiger reserves.
An illustration of the above simple solution is shown here for a quick reference. This solution is nascent and can be developed further to solve extended use cases of helping with the battle against Lantana Camara.
The illustration uses the following:
- Keras model that classifies weeds.
- Streamlit library is used to create a bare-bones UI that takes an image input and runs the AI model to identify the presence of Lantana Camara in the provided image.
- The streamlit application can be containerized and deployed on any Cloud.
For this illustration, a bare bone streamlit application UI is shown as below. The application performs image recognition of an uploaded image using the trained AI model.
- The image one wants to analyze is uploaded.
- The application loads the model, resizes the image and run the inference on the image.
- The inference is printed on the UI.
- Lantana is detected from the uploaded image.
- Clarity, lighting and other natural conditions pose difficulty in identifying Lantanas correctly from the images taken. A performant, accurate and well trained model makes the application so much more useful.
- The image of any other weed in the forest is uploaded to illustrate the classification.
- The image is rightly classified as “Not Lantana Camara”.
The illustration above was a simple example of the possibility of using AI in the detection efforts of Lantana Camara. There is a lot of ongoing research on using more advanced Machine Learning and AI techniques to help with the detection of Lantana Camaras in challenging terrains.
Pixel and Object-Based Approaches to detecting Lantanas in forests are suggested. Techniques such as the following are explored and discussed.
- Exploratory Analysis of Lantana Presence Using Logistic Regression
- Pixel-Based Classification
- Object-Based Classification
The Lantana detection use case can be extended to use the information from the Lantana predictions/identifications to operate a robotic de-weeder to de-weed Lantanas from interior or humanly less accessible forest lands.
This article explored an environmental issue of an invasive weed that is destroying the ecological balance and whose resolution can be assisted by applications using AI.
Tim Berners-Lee, the man who gave us the world wide web once said -
You affect the world with what you browse.
In the context of this article I would take the liberty to modify the quote to conclude with the spirit of this article.
You affect the world with what you build and how you apply technology.
