The current solution for using motion sensor technologies and implementing them in farm monitoring systems is also not without several drawbacks that hamper their efficacy, especially for the large-scale and remote environments of the farming sector. Almost all the conventional motion sensors and CCTV cameras are specially designed to detect human movement or larger animal intrusions but are not efficient at detecting smaller vertebrate pests like rodents that are well known to cause massive losses through crop and stored produce raiding. This consequently results in huge losses since rodent infestations are normally undetected until extensive damage has been done. Moreover, the current surveillance systems require a constant supply of electrical power, which is not available in most of the rural and developed central region areas.

Another problem in current motion sensor systems is that they are not real-time responsive, or at least not fully automated. Most typical systems employ the human operator to watch monitors for the footage or to respond to alerts, which raises a great deal of labor costs and hampers the efficiency of farm surveillance. All these systems are post-incident systems, which implies that these systems work following an incident, not to prevent it. Also, the utilization of these energy resources is costly and raises the environmental costs of operation, thereby making these systems unsustainable in their current utilization.

Hence, the research question lies in finding a solution that would not only enhance the detection accuracy of relatively small targets like rodents but also employ sustainable forms of power in the process. The resulting gaps can be filled by the proposed solar-powered motion sensor system, which is capable of using renewable energy, up-to-date sensors, automation, and systems to monitor both large and small intrusions more effectively in terms of energy consumption.

The primary objective of this proposed study is to develop a solar-powered motion sensor system capable of recognizing rodent invasion on farms within the monitored area, thereby enhancing efficient monitoring and surveillance efforts while consuming energy efficiently. The proposed model uses clean energy, thus a strong contribution to sustainable energy solution and environment conservation practice that avoids degradation, clear view of the farm and no pollution.

The proposed study is guided by the following research question:

This study proposes a new concept of farm monitoring and surveillance through the use of solar-powered motion sensors coupled with computer vision and deep learning algorithms for rodent invasion detection. Compared to conventional farm surveillance systems, which are primarily meant to monitor human interference or large animal movement and require continuous electrical power, the proposed system is intended to be energy-autonomous and is to be powered through solar energy. This not only provides for the need to power such devices in areas where there is no physical access to an energy grid but also provides a positive impact on worldwide sustainability as the need for fossil fuels lessens.

Furthermore, through the integrated feature of computer vision technology, it also has the possibility to differentiate between varying cycles of motion, for instance, to differentiate the wind-blown leaves with rodent movements. The use of deep learning in the algorithms helps the system to update the information that is collected, enhancing detection results in the future. This adaptive capability greatly improves the chances of the system to detect the smaller pests that are ignored in most traditional systems. Further, notification through a mobile application that reflects the actual situation in the field enables farmers to act instantly and avoid further harm that may require manual intervention.

This study’s uniqueness is found in the integration of renewable energy sources, advanced detection systems, and automation in the monitoring of farms. This approach has the possibility of radically changing the practice of monitoring farms, providing one distinctive tool for the farmers, especially those who are located in areas where they have no access to the normal electrical power source.

Figures 1 and 2 comprise the system infrastructure based on a motion sensor, flood light, motor, battery, speaker, rotating knob, a smart camera, a central database system and a solar panel.

3.2.1 Motion sensor

Image-based motion sensors analyze changes in the captured video to visually identify movement using cameras and computer vision algorithms ( Futagami et al., 2020). They are extremely accurate for this system because they use sophisticated algorithms for motion pattern detection and object tracking.

3.2.2 Data transmission protocol

WI-FI data transmission protocol facilitates data transfer between the system and the mobile application hosting the system’s GUI and database. WI-FI allows the transfer of data over a wide range ( Pahlavan and Krishnamurthy, 2020), hence most convenient for this type of technology. When the system detects motion and the camera takes a video, and processes it, it will use WI-FI to notify the farmer of the intrusion.

3.2.3 Rotating knob

The rotating knob moves the entire surveillance equipment, allowing the smart camera to capture images from all angles within the monitored area. It is connected to a motorized rotational mechanism, which rotates the entire surveillance equipment horizontally. Users can rotate the knob manually to adjust the viewing angle of the surveillance equipment.

It is powered by the battery, enabling the motorized rotational mechanism and motion detection sensors to pause temporarily when motion is detected. Therefore, ensuring that the camera focuses on the detected movement to capture images or video footage.

3.2.4 Smart camera

According to Kurniawan and Sofiarani (2020), a smart camera combines complex image sensors and processors with the capability to process images on its own without assistance from humans. It records high-quality video footage of the areas under observation only once the motion is detected, enabling remote surveillance, evidence collection, and in-depth analysis of any activity that is detected.

Smart cameras are equipped with high-resolution imaging sensors which record clear and detailed video footage. These sensors might use technologies like charge-coupled devices (CCD) or complementary metal-oxide semiconductors (CMOS), which provide better image quality even in low light. The camera works with Infrared sensor light to capture objects in low-vision light or during the night, ensuring that the model runs 24 hours.

3.2.5 Flood light

Floodlight produces light illumination in low-light or nighttime conditions. It typically uses energy-efficient LED (Light-Emitting Diode) technology and it is switched on automatically once the motion sensor detects an object. LEDs deliver a high light output while using very little power energy, making them ideal for solar-powered systems where energy conservation is essential ( Pulli et al., 2015). The floodlight is connected to the solar-charged battery, enabling it to draw power from it and operate effectively at night. This component is particularly useful for farm surveillance, as it can deter potential intruders and help the smart camera identify any activity occurring after dark.

3.2.6 Battery

The battery serves as the energy storage component in the system, ensuring a reliable and continuous power supply for the motion sensor, floodlight, and any other associated components. The preferred battery type for this system is lithium-ion batteries because they can be frequently charged and discharged without experiencing significant degradation and can be used for extended periods of time in solar-powered applications ( Manthiram, 2017).

3.2.7 Speaker

The speaker is the audio output device within the surveillance system. Its main purpose is to mimic sound, allowing for alarming and alerting. According to Bernardini, Bianchi and Sarti (2023), speakers utilize transducer technology to change electrical signals into sound waves, which are then released by speakers.

The motion sensor and speaker are combined so that when motion is detected, the motion sensor sends a signal to close the circuit linking the speakers. This causes the speakers to activate and emit sound alarms to frighten away intruders.

Figure 3 below is a flowchart showing the integration of system components. It also illustrates the flow of events and processes within the model.

In order to understand the proposed architecture, a simulation of image recognition was carried out using an image detection and classification pipeline to train and simulate the rodent recognition model. Data for training the model was obtained from YOLO classes obtained from COCO dataset. The data was grouped into five categories, namely: rat, mouse, squirrel, chipmunk and mole. The five categories formed the classes of the rodent recognition model.

3.3.1 Image recognition process in OpenCV

Preprocessing techniques are first used to simplify and lower noise in the image, as shown in Figure 4. These techniques include first grayscale conversion and then Gaussian blurring. Thresholding is done to create a binary image that highlights the important objects. This binary image has contours that show the edges of distinct objects. Each contour’s area is computed, and contours having areas outside of a predetermined range are removed. For each remaining contour, bounding rectangles are generated in order to identify and pinpoint each particular object in the image ( Duwal and Tamang, 2024).

The proposed solar-powered motion sensor system ( Figure 5) operates autonomously by harnessing clean, renewable energy from the sun. The photovoltaic (PV) module is the cornerstone of this system, converting sunlight directly into electrical energy through the photovoltaic effect. A photovoltaic system is an array of PV modules that comprise a number of solar cells that generate electrical power.

When sunlight, basically composed of photons, strikes the solar cells, the photons transfer their energy to electrons within the semiconductor material, hence exciting them ( Vinod, Kumar, and Singh, 2018). This energy transfer is the initial step in converting solar energy into electrical energy. The photon absorption circuit is shown in Figure 6.

Each solar cell contains a junction between two types of semiconductor materials: n-type and p-type. The junction between the n-type and p-type materials forms an electric field, causing the excited electrons to migrate to the p-type layer and leaving behind a static positive charge. Simultaneously, the holes wander across the junction, leaving behind a static negative charge. Eventually, a depletion zone forms at the junction, preventing further movement of charge carriers ( Kirchartz and Rau, 2018).

The separated static positive and negative charges establish an electric field across the depletion zone. This field generates the voltage necessary to drive current through an external circuit. As the semiconductor continuously absorbs sunlight, the energy excites more electrons, causing them to jump to the conduction band and leave behind holes in the valence band ( Chaudhery Mustansar Hussain, 2018). These freed electrons contribute to the electric current, moving toward the negative end, while holes move toward the positive end.

According to Chaudhery Mustansar Hussain (2018), when the absorbed photon energy exceeds the PV cell material's bandgap energy, the atoms in the semiconductor collide, freeing more electrons and generating an electric current. This process ( Figure 7) effectively converts sunlight into electrical energy, harnessing solar power for practical use.

A Solar Charge Controller performs several vital functions to optimize performance and safeguard equipment. Yassine and Anderson (2020) state that it prevents overcharging by regulating the voltage and current supplied to the batteries during sunlight hours, preserving battery life and preventing damage. Additionally, it blocks reverse current flow from batteries to panels, ensuring energy generated by the panels doesn't drain back into the battery during low-light or nighttime conditions.

This type of battery is predominantly used in portable electronic devices, therefore very convenient to be used in the model to power the devices in the farm and server station ( University of Washington, 2020). The Li-ion battery in the model consists of the anode, cathode, electrolyte, separator, positive current collector and negative current collector. Additionally, it has a charge-meter which displays the charge and discharge levels of the battery. The anode is the negative electrode that stores lithium and releases Li-ions during discharge whereas the cathode is the positive electrode that stores and releases Li-ions during charging. The electrolyte is liquid medium that allows for the flow of Li-ions from one electrode to the other. The anode and cathode are separated by a separator which allows free flow of Li-ions from one electrode to the other (US Department of Energy, 2023). Additionally, it prevents flow of electrons within the battery structure. The positive and negative current collectors act as the positive and negative terminals respectively ( University of Washington, 2020) ( Figures 8, 9 and 10).

Charging:

Discharging:

YOLO (You Only Look Once) is an image detection and categorization algorithm that was first developed by Redmon et al. in 2015 ( Ali and Zhang, 2024). It uses a Single-Stage-Object Detection mechanism to detect real-time objects in a single-short through the network. This mechanism simultaneously performs region proposal and object classification, hence streamlining the image detection process.

Several enhancements have revolutionized the YOLO frame since its inception as YOLOv1 to YOLOv11. The difference in the framework versions is based on enhanced accuracy, efficiency, and speed in detecting real-time data. This is due to a reduction in latency over each advancement, hence an improvement in accuracy ( Ali and Zhang, 2024). Figure 11 below illustrates the general architecture of a YOLO SSD detector.

Table 1 below summarizes the YOLO version and year of their establishment.

YOLO performs object detection by dividing an image into a grid and simultaneously predicting both class probabilities and bounding boxes. A deep-learning Convolutional Neural Network (CNN) performs feature extraction to generate these class probabilities and bounding boxes, through the convolutional layers. Detection of varying object sizes is enabled through anchor boxes at various scales. Finally, Non-Maximum Suppression (NMS) filters out low confidence and redundant predictions to refine the final predictions ( Jiang et al., 2022). Therefore, YOLO is a reliable and highly efficient framework to be used in our model for animal detection.

FPS shows system’s processing speed in real-time. X-axis shows the frame number from time to time while Y-axis shows the number of frames every second. The results display a high FPS (>20), indicating smooth running of the system, hence convenient for real-time monitoring.

This shows the type of detected animals together with their frequencies. From the results in Figure 13, the total number of animal detection was 49, Mice have the highest frequency of detection (19 counts) while Moles, Rats, Squirrels and Chipmunks have a count of 8,7,7,8 respectively. Farmers therefore, use the data as a guide in decision-making, assessing what measures to implement to prevent intrusion.

The Classification and Confidence Distribution show the confidence score of the system in perform animal classification. According to the results as seen in Figure 14, most detections scores 0.8-0.85, with the highest frequency recorded at 10 while lowest frequency at 1. A detection score of >0.8 justifies that the system is very reliable.

X-axis shows the frame number while the Y-axis shows the number of animal detections per frame. The Peaks show that there is a period of high animal activity; valleys show a reduction in animal activity or total absence; consistent line in frame progression shows a steady or constant animal activity; whereas spikes show sudden or irregular movement of animals into the frames. According to the results in Figure 15 above, the peak recorded 3 detections; most consistent detection of 2; low activity at 1; and absence of activity at 0. From the results, it is evident that the model is realistically applicable to capture both scenarios of rodent invasion and absence.