A Review on Recently Developed Technologies for Weed Recognition and Herbicide Application based on Digital Image Processing

Pankaj Malkani, K Asha, Atish Sagar, Abhinav Dubey, Akshay Kumar Singh, Prashant SinghInternational Journal of Current Microbiology and Applied Sciences 2019 · DOI

TL;DR

This review paper explores the exciting advancements in Site-Specific Weed Management (SSWM) that leverage digital image processing for precise herbicide application. By enabling machines to "see" and differentiate weeds from crops, these technologies promise significant reductions in herbicide usage, leading to environmental benefits and cost savings.

What Problem Does This Paper Address?

Weeds are a major adversary in agriculture, fiercely competing with crops for essential resources like water, nutrients, and sunlight, thereby drastically reducing yields. Traditional weed control methods, including manual labor, mechanical tools, and broadcast chemical spraying, each have their drawbacks. Manual methods are labor-intensive and costly, mechanical methods are often limited to inter-row applications, and broadcast chemical spraying, while effective, leads to widespread contamination of soil, water, and the environment, posing risks to human and animal health. The paper addresses the need for a more targeted and environmentally conscious approach to weed management, specifically by exploring how digital image processing can enable precision application of herbicides.

Methods

The core of the technologies discussed revolves around a three-step process for Site-Specific Weed Management (SSWM):

  1. Image Acquisition: This involves capturing real-time images of the field using various types of cameras. The paper highlights the use of RGB cameras (common digital cameras) for their affordability and ease of interpretation, as well as multispectral and hyperspectral cameras for more detailed spectral information that can aid in distinguishing between crops, weeds, and soil. Ground-based optical imaging is preferred for its higher spatial resolution compared to satellite imagery for smaller areas.
  2. Digital Image Processing for Crop-Weed Discrimination: This is the "brain" of the system. It involves several stages:
  3. Pre-processing: Cleaning up the acquired images by removing noise, resizing, and enhancing them.
  4. Vegetation Segmentation: Identifying regions in the image that contain vegetation, often using indices like the "excessive green index" or "normalized difference index."
    • Feature Extraction: Analyzing the segmented regions to extract relevant characteristics, such as morphology, texture, and spectral properties.
  5. Classification: Using algorithms, including machine learning techniques like Artificial Neural Networks and Principal Component Analysis, to classify these features and definitively identify weeds.
  6. Herbicide Application: Based on the weed identification and location determined by image processing, a smart spraying system precisely applies herbicides. This typically involves a microcontroller that receives information from the image processing unit and actuates solenoid valves to direct spray nozzles only at identified weed patches. Global Positioning System (GPS) is often integrated to provide accurate spatial coordinates for application.

Datasets

Not discussed in the paper. The paper focuses on reviewing existing technologies and their methodologies rather than presenting new experimental results based on specific datasets.

Key Insights and Ideas

  • The Power of "Seeing": Digital image processing transforms agricultural machinery from simple sprayers into intelligent systems capable of discerning individual weeds. This "machine vision" is the fundamental enabler of precision agriculture for weed control.
  • Significant Herbicide Savings: The consistent theme across various studies reviewed is the substantial reduction in herbicide usage, often ranging from 40% to over 75%, when compared to conventional broadcast spraying. This is a critical environmental and economic benefit.
  • Beyond Simple Detection: The sophistication of the image processing goes beyond just detecting greenness. Techniques like feature extraction (morphology, texture, spectral properties) and machine learning are crucial for accurately differentiating between crops and weeds, especially in complex scenarios.
  • Integrated Systems are Key: The success of SSWM relies on the seamless integration of sensors (cameras), processing units (computers, microcontrollers), and application mechanisms (sprayers, solenoid valves), often mounted on autonomous or semi-autonomous vehicles.

Key Results

The paper reviews numerous studies demonstrating the efficacy of SSWM technologies:

  • Herbicide Savings: Studies report savings ranging from 20.6% (Tangwongkit et al., 2006) to as high as 89.48% (Ozluoymak et al., 2018) in specific applications. Other notable figures include 48% (Tian et al., 2000), 66% (Gonzalez et al., 2016), and 75% (Shirzadi et al., 2013).
  • Weeding Efficiency: High weeding efficiencies are reported, with some systems achieving over 90% (Shirzadi et al., 2013).
  • Response Time: The systems are designed for real-time operation, with reported response times as low as 0.037s (Tian et al., 2000) and 0.05±0.003s (Zaman et al., 2011).

Limitations and Future Work

  • Commercialization Challenges: While successful prototypes exist (e.g., from Eco Robotix, John Deere), they are not yet widely commercialized. Higher initial development costs and the complexity of handling multiple crops with diverse weed types are cited as reasons.
  • Multi-Crop and Variable Weed Geometry: Addressing the challenges of identifying and managing multiple crop types with different weed species, and adapting to variations in weed geometry and characteristics, remains an area for future development.
  • Cost and Accessibility: The paper implicitly suggests that the cost of these advanced systems is a barrier to widespread adoption, especially for smaller farms.

The "Way forward" section emphasizes the need to overcome these hurdles to make these technologies more accessible and effective for a broader range of agricultural scenarios.

Summary updated

Abstract

Weed competes with a crop for nutrition, soil and water and, reduces its yields drastically. Conventional methods i.e. manual, mechanical and chemical mean have some limitations in controlling weeds. With the advancement in electronics and computers, Site-Specific Weed Management (SSWM) can provide a solution for precise weeds management. SSWM technologies with basic components and functions are described. SSWM consists of 3 basic processes i.e. image sensing, crop-weed discrimination, and chemical application. Digital image processing plays a crucial role in crop-weed discrimination and fascinates chemical applications.
BibTeX
@article{malkani2019review,
  author = {Pankaj Malkani and K Asha and Atish Sagar and Abhinav Dubey and Akshay Kumar Singh and Prashant Singh},
  title = {{A Review on Recently Developed Technologies for Weed Recognition and Herbicide Application based on Digital Image Processing}},
  journal = {International Journal of Current Microbiology and Applied Sciences},
  year = {2019},
  doi = {10.20546/ijcmas.2019.812.077},
}
Figure