Crop Load Estimation
Accurate and early estimation of citrus yields when the fruit is green is important for both producers and agricultural cooperatives to be competitive and make informed decisions when selling their products. Yield estimation is key for predicting stock volumes, avoiding stock ruptures and planning h
2025-06-28 16:31:00 - Adil Khan
Crop Load Estimation
Project Area of Specialization Artificial IntelligenceProject SummaryAccurate and early estimation of citrus yields when the fruit is green is important for both producers and agricultural cooperatives to be competitive and make informed decisions when selling their products. Yield estimation is key for predicting stock volumes, avoiding stock ruptures and planning harvesting operations. Visual yield estimations have traditionally been employed, resulting in inaccurate and misleading information. Therefore, the main goal of this project is to develop an automated image processing methodology to detect and count citrus fruits on individual citrus plants using deep learning algorithms. Computer vision and machine learning will be applied to automate fruit counting. First of all, a custom dataset will be prepared (already done) from AARI Research Orchards (31.389, 73.004) of citrus plant. The collected dataset will be annotated (already done) using the annotation tool Labellmg. After that a “faster_rcnn_inception_resnet_v2_640x640” deep learning model will be trained on custom dataset using Nvidia: Jetson Nano Developer Kit B01 to count oranges in the obtained images. Finally, the trained model will be deployed on cloud platform and connected to the mobile app will be created that takes the picture as input and estimates the citrus load in the orchard. For expert opinion we got help from co-supervisor Prof Dr. Muhammad Jafar Jaskani at Institute of Horticultural Sciences.
Project ObjectivesA mobile application will be created that will take images of citrus plants and count the fruit on the plants, that will speed up the current human-based sampling protocol. This will help the farmers, buyers and suppliers in the following ways:
- It will impact on decisions about orchard management and value chain activities.
- It will help in planning the budget,
- Pickers and packing shed
- All the organization around managing the crop.
- Data collection method:
We took 1000+ pictures of immature and mature oranges from AARI Research Orchards (31.38913197028696, 73.0042145886367) of different plants for detecting and monitoring the citrus plants at different stages e.g., after flowering when the fruit was immature, pre-mature and mature, to find out how much fruit falls and to keep track of final count.
Citrus plants (31.38913197028696, 73.0042145886367)
- Annotation:
We performed a manual labelling process to identify where the fruits are in each image using LabelImg tool.
- Detection Model:
For counting of the fruits, we will use object detection algorithm “faster_rcnn_inception_resnet_v2_640x640”. The algorithm can accurately and quickly predict the location of different objects in the image. It takes 0.2 seconds to test each image; it can even be used for real-time object detection. This model will be trained on Nvidia: Jetson Nano Developer Kit B01.
- Deployment:
The trained model will be deployed on cloud platform and connected to the mobile app, will be created that takes the picture as input and estimates the citrus load in the orchard.
- Crop load will impact on decisions about orchard management and value chain activities. It is recommended that both early and late estimates are made each season.
- The early estimate will help to determine whether practices such as hand thinning are necessary, while the latter one provides an estimated yield for the season.
- Yield information is important for discussing supply capacity with buyers as well as organizing orchard management and packing shed logistics at the beginning of the season.
- This will help to monitor tree production and its management for later years.
The final deliverable is a mobile application that takes a picture of the tree and counts the total number of fruits in that tree and then using this counting predicts the total number of fruits in that orchard.
Final Deliverable of the Project HW/SW integrated systemCore Industry AgricultureOther IndustriesCore Technology Artificial Intelligence(AI)Other Technologies Cloud InfrastructureSustainable Development Goals Zero Hunger, Decent Work and Economic Growth, Responsible Consumption and ProductionRequired Resources| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 78600 | |||
| LG G5 Compact 360 Degree Camera | Equipment | 1 | 34250 | 34250 |
| Nvidia: Jetson Nano Developer Kit B01 - 4GB | Equipment | 1 | 34500 | 34500 |
| Data Collection and Annotation | Miscellaneous | 1 | 9850 | 9850 |