This section independently discusses the findings of each repellent system based on the four divisions outlined in the methodology.
Animal detector and identifier
Two main detection approaches have emerged from research. The first method involves a single detector activating a controller or actuator upon animal detection, typically using ultrasonic, infrared, or resistance sensors. The second method uses sensors to trigger central components, such as activating an identification system camera for subsequent image processing. Within deep learning research on object detection, two main architectural categories were identified: one-stage and two-stage algorithms. The one-stage category includes well-known members like YOLO (You Only Look Once), Single Shot Detector (SSD), Detect Net and Squeeze Det. The two-stage category encompasses Region-based Convolutional Neural Net-work (R-CNN), Faster R-CNN, Feature Pyramid Network (FPN) and Region-based Fully Convolutional Network (R-FCN)
(Du et al., 2020). One-stage detectors prioritize high inference speeds, while two-stage detectors excel in localization and recognition accuracy. The investigation explored standard algorithms and descriptors widely recognized in machine learning and computer vision, such as Histogram of Oriented Gradients (HOG), Scale-Invariant Feature Transform (SIFT), Speeded Up Robust Features (SURF) and Oriented FAST and Rotated BRIEF (ORB), for object identification. The choice of a specific approach depends on the research objectives, conditions and characteristics of the study subject. Research findings underscore the effectiveness of the YOLO architecture for real-time animal detection, particularly in scenarios where animals are not small or densely embedded in the background
(Abed and Murugan, 2023). Various studies employed resistance sensors, passive infrared (PIR) sensors and ultrasonic sensors for detection purposes. For example, resistance sensors combined with fencing wire detected animal presence by closing an open circuit upon contact
(Deshpande, 2016). PIR sensors detected infrared radiation emitted by external objects, successfully identifying various objects using an HC-SR501 sensor
(Priyadarshini et al., 2015). Ultrasonic sensors enabled distance-based detection, triggering cameras when animals approached within 3 meters of the farm. After processing images with ORB and MATLAB, it was easier to identify specific animals based on their unique features, with an impressive 82.5% success rate
(Sharma et al., 2017). Similarly, other studies employed the HARR cascade classifier descriptor, resulting in a detection accuracy of 78.1%
(Sharma and Shah, 2017). Motion detection via ultrasonic sensors was enhanced with GPS technology, enabling real-time location tracking of animals through IoT and the Ubidots application
(Manoharan et al., 2020). In some instances, PIR sensors were coupled with cameras to detect motion, capture still im-ages and record videos, providing immediate notifications
(Bavane et al., 2018; Pooja and Bagali, 2016 To facilitate video uploads to cloud platforms like Dropbox, Bash scripting was employed in certain cases, ensuring data persistence and accessibility. Additionally, the integration of RFID labels helped distinguish between authorized and unauthorized users, enhancing security
(Ross et al., 2020). Nevertheless, the system exhibited misclassifications, raising concerns about its practical efficiency in dynamic environments. While fuzzy logic and convolutional neural networks achieved high accuracy in animal identification, they rely on computer system support or cloud storage, which may not be universally accessible
(Mohammed and Hussain, 2021).
Control system and embedded board
The control system serves as the brain of the setup, overseeing the interaction of devices and sensors for seamless integration of data collection, processing and communication, enabling real-time monitoring and control. Versatile microcontroller units (MCUs) like Raspberry Pi, Arduino boards and ESP32 are often used as central control hubs, facilitating integration with various sensors and cameras. Communication between components occurs through various means, including GSM modems operating within a 2 G signal, using SIMCOM SIM300 SIM cards for message transmission and remote object management
(Deshpande, 2016; Rekha et al., 2017 a) . A Tri-band GSM/GPRS modem is commonly employed, with default frequencies set at EGSM 900 MHz and DCS 1800 MHz. Researchers have developed Android applications to collect real-time data from the farm, enhancing the precision of Wireless Sensor Network (WSN) systems for evaluating sensor deployment requirements
(Rekha et al., 2017b). The Raspberry Pi board has played a crucial role in centralizing system components, providing automatic notifications to farmers
via SIM900A modules and employing advanced image and video detection algorithms to distinguish between human and animal intruders
(Bavane et al., 2018; Pooja and Bagali, 2016). Furthermore, certain designs have been executed with a broader integration of sensors to enhance the effectiveness of smart farm management
(Durga et al., 2017). Fig 2 and 3 visually represent the block diagram encompassing the sample systems, illustrating different design connections and components.
Repellent actuators
Repellent systems typically incorporate one or more of the following elements to deter intruders: I. Sonic Buzzer; II. Ultrasonic Buzzer; III. Flashlight; IV. Odor spray; V. Firecracker; VI. Smoker; VII. Manual. The efficacy of hearing-based mechanisms in repellents hinges on their design and calibration within the auditory range. Auditory perception is influenced by the spectrum of unpleasant frequencies, which varies depending on the biological characteristics of each organism. Humans can perceive frequencies from 20 Hz to 20 kHz, with the hearing threshold set at 0 dB as the reference level. Ultrasound includes sounds beyond 20 kHz, detectable by certain species. Creatures exhibit hearing abilities across the sonic, ultrasonic and infrasonic ranges, with some capable of perceiving sounds in all three domains. Table 1 compiles data from various studies on the hearing ranges of different animals
(Heffner et al., 2020 a;
Heffner et al., 2020 b;
Jakobsen et al., 2021; Menda et al., 2019; Morley et al., 2014; Trevino et al., 2019).
Repellents that utilize unpleasant frequencies often manipulate the amplitude of sonic and ultrasound waves. For instance, some systems employ a sonic buzzer
(Deshpande, 2016;
Sharma et al., 2017). Another system, presented by
Yusman et al., (2018), by emits ultrasound in the 25 to 40 kHz range it successfully repels civets, cows, goats and monkeys. However, its limited detection range of 5 meters falls short of real farm applications. Studies have identified the frequency range of 25 kHz to 65 kHz as effective for repelling most pests, with specific frequencies between 38 kHz and 44 kHz for deterring
(Saini et al., 2016). However, a recurring issue with ultrasonic-repellent devices is that pests can adapt to the frequency over time, rendering them ineffective. To address this, ultrasonic pest-repellent devices, as conceived automatically switch frequencies every 30 seconds, preventing pests from adapting to the ultrasonic waves. Emitting modulated frequencies ranging from 25 kHz to 65 kHz, these devices induce distress and discomfort in pests. However, they necessitate advanced calibration and monitoring, limiting their widespread applicability
(Nair et al., 2017). An alternative approach to repelling rodents involves the use of scents and plant products to stimulate their olfactory sense instead of rodenticides, offering a different means of repelling pests. Experiments with common voles (Microtus arvalis) involved T-maze trials to assess their aversive responses to various odors for repelling these rodents
(Schlotelburg et al., 2019).
Certain researchers integrated repellent systems into their designs, combining a buzzer, rotten egg sprays, electronic safe-firecrackers and focused light based on light intensity to deter animals from agricultural areas
(Pooja and Bagali, 2016).
Power supply
Power sources for the devices in the reviewed studies vary based on factors like capacity, portability, indoor/outdoor use and complexity. These power options include: I. Various battery types; II. AC power; III. USB computer cables; IV. Solar panels.
In many cases, these devices rely on DC power, typically under 12 volts. For instance,
Rashid et al., (2017) utilized a 10-watt solar panel alongside an LM317 voltage regulator to achieve the desired voltage; Fig 4 depicts the scheme of the project.
According to the design model for Ratspay, if four images are captured daily using AA lithium batteries with a capacity of around 3,000 mAh, it can operate for approximately six months. Alternatively, a single rechargeable LiPO cell can be used and recharged every six months for longer usage
(Ross et al., 2020). In an alternate setup, power was sourced from either an AC-to-DC converter or a battery, while some systems drew power from a computer
via a USB cable
(Yusman et al., 2018). The system boasts advantages like flexibility in power sources (Solar, AC mains, or batteries) and continuous monitoring; however, it is not without drawbacks. Challenges include the constant need for battery charging checks and the complexity of large-scale deployments
(Deshpande, 2016). Intelligent, clean energy-powered devices play a vital role in deterring intruders in agricultural fields, aligning with Sustainable Development Goals (SDGs) 7 and 15. Reducing pollution of any sort, especially industrial
(Abed et al., 2019) and agricultural, is an inevitable solution to preserve the land and create a platform for survival.
A review of major studies and projects highlights the critical importance of preventing damage from trespassing animals, driving investigations in various fields and utilizing diverse techniques. Numerous studies have focused on modifying and enhancing specific components of repellent devices, yielding valuable results. Additionally, unrelated research has provided constructive solutions to enhance existing designs. While most laboratory studies and prototypes have centered around pests or rodents, there is a scarcity of applied research on a farm scale. In summary, Table 2 provides a comparative overview of findings from distinct articles. Equipment selection for projects depends on factors such as invasive species, plant characteristics, environmental conditions, device cost, sensitivity to protection and coverage area.
This paper explores strategies for repelling unwanted animals from farms, emphasizing Rajasthan’s unique farming conditions. Solar energy, batteries and direct current are essential power sources, with solar energy being practical in sunny regions, eliminating the need for frequent battery replacement. Arduino and Raspberry Pi boards are prominent choices for control segmentation. Arduino simplifies sensor interfacing, running on battery power with onboard storage, while Raspberry Pi, resembling a mini-computer, presents complexities when battery-powered and relies on SD card storage. Intelligent security technology aims to reduce human intervention while requiring human oversight. IoT-enabled systems facilitate remote device control and monitoring, offering insights into farm conditions and invasive species. This device aims to enhance crop protection, aligning with SDG indicator 2.4.1. To foster sustainable agriculture and resilience engineering, designing intelligent devices utilizing environmentally friendly renewable energy sources for intruder control is essential. Table 3 provides a device configuration for sustainable crop protection in Rajasthan, based on insights and specific requirements.
