Estimation of Fruit load on a Tree using Machine Learning

Project Title

Estimation of Fruit load on a Tree using Machine Learning

Project Area of Specialization Artificial IntelligenceProject Summary

Pakistan is an agricultural country that produces numerous types of crops, vegetables, and fruits. The fruits are a major part of our agricultural produce. According to the latest report by the Ministry of National Food Security & Research of Pakistan (Dec. 2021), Pakistan produces more than 29 types of fruits including citrus, mango, apple, guava, peaches, and apricot, etc. These fruits are not only consumed domestically but also exported to other countries. According to the latest figures for 2019-20, an area of 711274 hectares is under fruit cultivation, and the total production of different fruits is 7126275 tons. Whereas the production of the four topmost fruits Citrus, Mango, Banana, and Apple is approximately 4736101 tons (66.46% of total production). Most of the data related to the estimation of fruit production are collected and compiled by the provincial Crop Reporting Services through surveys of the fields. A major concern in this regard is an early and accurate estimation of fruit production when the fruits are still on the trees. This estimation plays a vital role in the overall planning regarding the decision related to the local supply and export of different fruits as well as the overall agricultural growth of the country. In addition to the national level estimation of the production, an individual or local level estimation of the fruit production is also very important. In Pakistan most fruit growers sell their orchards fully are partially at or before the time of fruit plucking. The buyer of the orchard makes arrangements to pluck the fruit from the trees and then transport it to the market. At the time of selling/buying both the parties depend on human experts for the estimation of fruit load (weight and count) on individual trees and in the orchard as a whole. An accurate estimate of the fruit is very important for both parties, not only for the agreement on the price but also for making arrangements for plucking, packing, and transportation of the fruit. In this project, we are planning to develop a Machine Learning based mobile app that can provide a good estimate of the fruit load on a single tree by taking multiple pictures of the tree from different directions including aerial views. After obtaining the average estimation of fruit load on different trees we can extend it to find out the fruit load of the whole orchard. The server-side part of the project will keep a record of all the registered mobile app users and provide different functionalities as and when required by the user such as fruit load estimation calculations. Initially, the system will be developed for the estimation of the fruit load of Kinnow, however, the same system can be extended for other fruits such as Apple, Mango, Apricot, etc.

Project Objectives

The following are the main objectives of the project;

• Development of a machine learning-based model that can estimate the fruit load of a specific fruit using tree/plant images with at least 80% accuracy. In our FYP we have very limited time so we will develop the model only for the estimation of fruit load for the Kinnow.
• Development of an android based mobile app that can be used in the field to collect data/images of the trees. The app should also be able to keep a detailed record of the user. The app should also be able to work offline for a quick response.
• Development of a server system to collect the data from mobile app users and provide fruit estimation results to the connected users if required. The server side should also be able to keep the record of all the registered users.

Project Implementation Method

This project consists of three major modules or parts (i) The building of an ML-based fruit estimation model using image data (ii) the development of android based mobile app (iii) the server-side that processes the queries and keeps the record of the mobile app users. These three parts require different implementation methods, discussed as under

1. The ML-based fruit estimation model will be built using the image data collected from different orchards of Kinnows. Four to five images of each tree with fruit will be taken from different directions including an aerial view. These images will be processed/segmented and the area covered by the fruit will be determined based on the color and shape of the objects. The actual fruit load on the tree will also be recorded while fruit plucking from that tree. The actual fruit load will be used as output and the images of the tree will be used as input to the ML algorithm. Different selected ML algorithms will be used for estimation purposes. These algorithms will be selected after a thorough survey of the literature. The final algorithm will be selected based on performance criteria i.e. Accuracy or F-measure. The best-suited algorithm will be fine-tuned for better results before the final decision.
2. The android based mobile application will be developed for the user in the field. The app will be able to take the images of the fruit tree directly and by using the camera installed on a drone.  This app will also keep records related to different fruits, seasons, and years of the registered user with total area, the number of trees, and different types of fertilizers and pesticides used. This app will also interact with the server for providing information and fetching estimation results as and when required.

The server-side software will keep a record of all the registered users. Process the image data received and provide estimation results and other information required by the user.

Benefits of the Project

The following are the expected benefits of the project 

•  Help the farmer to get the estimated fruit load of their orchards before plucking the fruit.
• Help buyer to get estimated fruit load of the orchard before buying it.
• Help grower/buyer with making arrangements regarding fruit plucking, packing, transportation, etc. well before time.
• Minimize the dependence on human experts to obtain an estimation of the fruit load on the trees.
• Keep a record of different orchards, seasons, and years for different fruits.
• Provide tehsil, district, and province-level estimation of fruit with minimum efforts on field surveys.
• Continuous improvement in the accuracy of estimation of fruit load using actual figures after the fruit is plucked.

Technical Details of Final Deliverable

The final deliverable consists of two software modules (i) Server-side application and (ii) android based mobile application.

1. The development of the mobile application is based on the following frameworks

• Flutter
• Firebase
• FirebaseDatabase
• Rest Full API
• YOLO v3

2.  The server-side application will be based on a mixture of the following technologies and frameworks. We will select the more appropriate technology/framework as and when the development progresses

• HTTP
• Django
• HTML
• JSON
• XML
• Python
• YOLO v3

Final Deliverable of the Project Software SystemCore Industry AgricultureOther Industries Food Core Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Responsible Consumption and ProductionRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	80000
Drone Camera	Equipment	1	70000	70000
Stationary	Miscellaneous	1	4000	4000
Traveling	Miscellaneous	1	6000	6000