Machine Vision Based Autonomous Weed Detection and Herbicides Spraying Robot

Project Title

Machine Vision Based Autonomous Weed Detection and Herbicides Spraying Robot

Project Area of Specialization Electrical/Electronic EngineeringProject Summary

According to the Pakistan Bureau of Statistics (PBS), the average annual growth rate of Pakistan is 2.69%. This growth rate indicates that the future food needs of Pakistan will be increased which causes the food crisis. Therefore, we must take the necessary steps to make farming more efficient, which can be achieved with precision farming. Pakistan Agriculture Research Council (PARC) stated that The main reasons for the low yield of wheat in Pakistan are uncertainty of rainfall, lack of water, substandard methods of cultivation, malnutrition, and weed invasion which may decrease up to 25-30 percent of yield, Weed control are the important aspect in precision agriculture, weeds are dangerous and unwanted plant, weeds grow close to valuable plants, thus feeding on mineral plants, weed control is one of the areas which demands automation. With standard weed control systems, herbicides are sprayed uniformly throughout the field. This procedure is incompetent as pretty much 20% of the shower achieves the plant and under 1% of the synthetic really adds to weed control, prompting wastage, sullying of nature and wellbeing issues in individuals [1],

These methods increase the crop production cost and not sustainable as well. the reduction of human labor and various agrochemicals that are used for weed reduction which can lead to sustainable agrarian culture. Robots that can perform targeted weed control offers the potential to contribute towards this issue, for example, through specialized weeding actions such as mechanical weed removal. With evolving technology in robotics and artificial intelligence, these problems can be overcome. A prerequisite of such robotic systems is a reliable and robust plant classification system that can distinguish crop and weed in the field. The key objective is to reduce the reliance on agrochemicals such as herbicides due to its side-effects on the environment, biodiversity, and partially human health. These systems must be able to locate the weeds in the field and as a result, the herbicide sprayers are instructed to spray directly on the desired areas. In this context, we propose an autonomous weed detection robot based on a machine learning vision system that identifies the unwanted weed and sprays the affected area. The proposed solution will be cost-effective and will reduce environmental pollution by reducing the number of herbicides. To identify weeds, different attributes have been used in recent papers. One of these attributes is color or spectral reflectance properties. To identify beets among different weed species, Feyaerts and van Gool [2] used a spectrograph camera, and up to 86% classification accuracy was reached, although because of using six narrow spectral bands, the scheme did not apply to in-field purposes.

Project Objectives

The objective of this project is to replace manual weeding in organic farming with a device working autonomously at the field level. Developing such a device is considered as a design problem. There are main following objectives,

1. Mechanical design of robot.
2. To identify weeds and crops by using machine vision.
3. To find which method is best to remove weeds with the most accuracy.
4. Precision spraying at the targeted weed .
5. Self-power production using solor energy.
6. To reduce herbicides investment in Pakistan.
7. To reduce labor intervention for harvesting in crops field.
8. To increase production of crops in Pakistan.
9. To reduce herbicides investment in Pakistan.
10. To overcome Pakistan food crisis.
11. To introduce precision farming in Pakistan.

Project Implementation Method

Computer Vision Based Weed Detection:

Computer vision approaches generally deal with utilizing complex image processing techniques to extract meaningful features from a given set of images. The general method for weed detection using computer vision starts with the acquisition of a digital image that will typically contain weeds mixed with crops, as well as soil in the background. Subsequent image processing aims to pinpoint the location of weeds, such that the system can swiftly and precisely guide the weeding mechanism. In some cases, the machine vision system is also designed to be used for guiding the navigation system of the robot. Computer vision approaches have been widely and historically used for post hoc weed identification [1] [2], but there is still progress being made on computer vision techniques for identifying weeds in the field in real-time for robotics applications.

Deep Learning in Weed Detection:

While computer vision approaches have been fairly successful in weed identification, in recent years deep learning models such as Convolutional Neural Networks (CNNs) have emerged as the dominating models in computer vision tasks. Image classification and object detection models based on CNNs have also been proposed for the task of weed discrimination, and successful implementations have been presented in multiple agricultural problems In case of weed detection, the problem becomes a special case of plant species classification. The advantage in using deep learning models is that they make segmentation and feature selection redundant since the extraction of features and the mapping of learned features to an output result are built into the network. Models trained on large datasets are used in cases of applications where the dataset may be limited, and this has been found to be advantageous even if the pre-trained model was trained on completely different data .The requirement for a large amount of training data is one of the challenges associated with using CNN models for detection of weeds. Like all supervised models, the object detection models are only able to perform well on data that comes from the population of data which it has already been trained on. A commonly used technique to mitigate this problem is data augmentation, where transformations such as rotated versions of the available images are used as additional data.   An additional challenge that arises with the large datasets for deep learning algorithms is the necessity to label the images. Labeling of data is also a problem with non-deep learning methods previously discussed and several studies have attempted to create unsupervised detection methods without the need to select or label training data.[3][4][5]

Benefits of the Project

As Pakistan is an agricultural country, Pakistan's economy depends upon agriculture. If agriculture grows, it will directly increase Pakistan's economy, It will affect Pakistan's economy by decreasing the import of herbicides. It will increase crop production in Pakistan. It will help the Pakistan government to overcome the food crisis. It will bring precision farming to Pakistan, It will not act as a separate load at the power grid. It will produce its own energy using solar panels.

Technical Details of Final Deliverable

The final deliverable of the project will be a robot powered by solar panels, which will be able to detect in row weeds using a camera then it will spray at the detected target using a controlled sprayer, image processing, sprayer control system, and weed detection model processing will be conducted using jetson nano developer kit.

Final Deliverable of the Project HW/SW integrated systemCore Industry AgricultureOther IndustriesCore Technology RoboticsOther Technologies Artificial Intelligence(AI)Sustainable Development Goals Zero Hunger, Good Health and Well-Being for People, Industry, Innovation and Infrastructure, Responsible Consumption and ProductionRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	74700
NVIDIA Jetson Nano Developer Kit	Equipment	1	25000	25000
Raspberry Pi Camera	Equipment	1	9000	9000
BLDC Motor	Equipment	2	6000	12000
Motor Drives	Equipment	2	1000	2000
DC Water Pump	Equipment	1	500	500
DC Solenoid Valves	Equipment	3	900	2700
Solar Pannel	Equipment	1	8500	8500
Power supply	Equipment	1	5000	5000
Robot Mecahnical Design	Miscellaneous	1	5000	5000
Miscellaneous	Miscellaneous	1	5000	5000