Empowering Mango Farming with AI

India has a strong agricultural tradition that strongly supports the country’s economy and its overall gross domestic product. Of these, the mango stands out as the most popular fruit, often referred to as the ‘king of fruits’, recognised for its taste, rich colour and a truly historic and traditional connection with the nation’s food culture.

Mango is the national fruit of India and the country contributes to nearly half of the global output of mangoes and is beyond 20 million metric tons per year. Mango contributes 18.4% to agricultural GDP and provides employment to more than 54% of the agricultural workforce; therefore, it is essential for food security: particularly Gujarat, Maharashtra, Uttar Pradesh, Andhra Pradesh and Karnataka. Being a export product that has gone past USD 860 million in its annual turnover, it helps in growth of economy and rural income.

However, similar to other producing countries, mango cultivation in India is threatened by factors such as pests and diseases, climatic change, price volatility of the produce and market and supply chain constraints. Some of these diseases include; Anthracnose, powdery mildew, and bacterial spot all of which result in low yields, poor quality tomatoes and reduced farmers income. The first thing about this condition is that you must get a check-up as early as possible, and the identification of the disease must be precise. Even the diseases on the mango leaves result to millions of dollar losses in the Indian mango business, further suggesting why effective solutions are wanted.

Moreover, diseases that affect the mango trees include, anthracnose, powdery mildew, bacterial spot and give symptoms different from one another; be it dark lesions, powdery appearance or water-soaked blemishes on the leaves, flowers, and fruits in all cases. Some of these diseases affect crops through pathogens, and environmental conditions such as humidity, temperature, rainfall and pests/insect can easily be noticed if the diseases are diagnosed early enough. However, the traditional approaches to visual inspection despite its importance are tremendously time-consuming and dependent on the experience of the observer, in addition to being fairly subjective especially when aiming at capturing minor or initial sign of deterioration. Automated disease detection systems give the clinician a faster, more accurate, and objective means of analyzing vast amounts of individual patient data in order to offer timely and accurate diagnoses. Since the systems facilitate timely intervention, they assist farmers prevent/contain diseases, minimise crop losses and encourage sustainable farming practices.

During a field trip to a mango farm in Surat, Gujarat, I had the opportunity to interact with Ronak Desai, a mango farmer with over 150 mango trees. He shared insights into the challenges of mango cultivation, emphasizing the detrimental impact of mango leaf diseases. Early prediction of these diseases, he noted, is crucial for effective treatment and minimizing harm to the cultivation process. However, manual prediction requires skilled labor and is prone to errors. When I introduced our project's mission to develop an AI-powered disease detection system, Mr. Desai expressed enthusiasm for its potential to revolutionize mango cultivation by providing accurate and timely disease prediction. 

Our proposed Artificial Intelligence driven mango leaf disease diagnostic tool plans on transforming how farmers diagnose and treat diseases within their mature mango trees. The system operates based on analyzing digital images of the mango’s leaves, extracting features related to different diseases on its own, and categorizing the images as healthy or diseased. This cuts out the need for visual inspection, and instead gives the farmers fast and accurate results. By design, the system is easy to use for farmers, even if extensions lack formal computer education and training.

Our system is based on the neural networks that resemble the structure of the human brain and are rather complex algorithms. These networks can find dependencies and correlations in a huge amount of data and that is why they are used in image recognition. In our system, neural networks are trained on a large number of mango leaves images to determine whether the given input is a healthy mango leaf or the diseased one with almost actual percentage. This capacity to learn and adapt makes neural networks a valuable tool in disease detection and it distinguishes our system through giving farmers timely and accurate information for the management of diseases. 

The MangoLeafBD was a very useful source for the given problem. It is a publicly available data, which includes 4000 high resolution images of mango leaves belonging to 8 classes- 7 diseases and healthy leaves. Mobile phone cameras were used to acquire the images to consider variable real-world conditions of mango orchards in Bangladesh. As shown in Table 2, we produced five sets of 500 images each, which contributes to establishing a balanced dataset to prevent the model from overemphasizing any pathologic class and enhance its precision.  

Although an ideal dataset should have been collected from India, MangoLeafBD-collected from Bangladesh was the closest. Since climatic conditions of fruits are highly correlated by geographical area, and Bangladesh has same climate as India having same type of mangoes used in the dataset is ideal for the papyrus project. Moreover, this dataset has not been adopted frequently in the past literature, thereby enabling the utilization of extended forms of machine learning models for the identification of mango leaf diseases.

Data Acquisition and Preprocessing: The MangoLeafBD dataset was collected by us assuming plenty of pictures of both healthy mango leaves and those influenced by diseases. Before feeding the images to the model all the images were normalized and all the images were resized for better results of the models.    

Model Selection and Training: We chose several CNN architectures such as MobileNet, VGG 16, Inception V3, DenseNet 121, AlexNet ResNet 50 and Efficient Net B7. The above mentioned models were trained with the preprocessed data set on the efficient optimization algorithms.  

Model Testing and Evaluation: The trained models were assessed separately using a testing data set. Model Comparison: In this process, we evaluated the performance of various models to determine the best structure for Mango LEAF disease identification and categorization.

Python: The core programming language for data analysis, model development, and implementation.  

Kaggle: A platform providing datasets, notebooks, and a collaborative environment for data science tasks.  

Google Colab: A cloud-based environment offering free access to GPUs for accelerated model training.  

Matplotlib: A versatile library for creating data visualizations.  

Keras: A high-level API for building and training deep learning models.  

TensorFlow: The backend framework for Keras, providing efficient execution of deep learning operations.  

NumPy: A fundamental library for numerical computing in Python.  

GPUs: Graphics Processing Units for accelerated model training.  

MobileNet: A lightweight and efficient CNN architecture.  

VGG16: A deep CNN architecture with 16 layers.  

Inception V3: A CNN architecture known for its efficiency and accuracy in image classification.  

DenseNet121: A CNN architecture with dense connectivity for improved feature reuse.  

AlexNet: A CNN architecture effective in image recognition tasks.  

ResNet50: A CNN architecture with residual connections for addressing the vanishing gradient problem.  

EfficientNetB7: A CNN architecture achieving high accuracy while maintaining computational efficiency. 

This project has the potential of transforming mango growing in that there will be a proper means of identifying diseases with high efficiency. Further improvement can be made to broaden the diseases which the system identifies and the development of the system into an easy to use mobile application for farmers. This article reveals further optimising the system enables to detect diseases and have real time reports to manage those diseases better. The Advancements proposed in the project integrate with the existing studies giving a relevant contribution to the AI sector in agriculture, providing sustainable solutions to enhance the production quantity and quality assurance.