MFPN-ASPP is a segmentation network for photovoltaic (PV) thermal images. It takes a thermal image of a solar installation and outputs a pixel-level binary mask that marks anomalous regions. In the Scientific Reports paper by Galal et al., published 8 October 2026, the authors report a Dice score of 0.8440 and an IoU of 0.7564 on a single dataset of 1,009 drone-captured thermal images from one solar plant in South Africa. These are measurements on that benchmark, not a guarantee of how the model will perform at another site.
What the name means
The title names four pieces. The task is defect segmentation: instead of labeling a whole panel image as faulty or healthy, the model assigns every pixel to either anomaly or background, which is what binary semantic segmentation means. MobileNet (here MobileNetV2) is the lightweight convolutional encoder that extracts features from the image. Feature pyramid refers to a Feature Pyramid Network (FPN) decoder that merges features taken at several depths of the network, so that both small and large anomalous regions are represented. ASPP, atrous spatial pyramid pooling, is a head that processes the merged features at several receptive field sizes to add surrounding context.
As an Amazon Associate I earn from qualifying purchases.
How the model turns a thermal image into a mask
- Normalize the thermal image. The paper’s processing flow starts by normalizing the input thermal image before feature extraction.
- Extract features with MobileNetV2. The encoder produces feature maps at several scales.
- Fuse the scales with FPN. The decoder combines those maps into one multi-scale representation.
- Add context with ASPP. The fused representation passes through the ASPP head.
- Output probabilities and a mask. The model produces a defect probability for every pixel. A probability threshold of 0.5 converts these into the final binary mask.
Inside the ASPP head
The ASPP head runs four parallel branches over the fused features:
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →- Three convolution branches with dilation rates 1, 6 and 12. A dilation rate spaces the filter taps further apart, so a filter sees a wider area without adding parameters. A rate of 1 is an ordinary convolution.
- A global average pooling branch, which summarizes the whole feature map into one context vector.
The branch outputs are concatenated and projected, and only then upsampled to the output resolution.
#1 Best Overall
- ⚡ Professional-Grade PV Testing Measures maximum power (Pmax) up to 1000W, open-circuit voltage (Voc: 12-80V), and short-circuit current (Isc: 35A) with ±0.8% accuracy, ideal for validating solar panel performance in R&D, manufacturing, and field maintenance.
- ⚡ MPPT Efficiency Optimization Tracks Vmp (80V) & Amp (35A) in real-time to identify panel degradation or shading issues, helping installers maximize energy harvest and ROI for residential/commercial systems.
- ⚡ Industrial Safety & Durability Rated CAT III 1000V/CAT IV 600V with double-insulated probes, meeting IEC/EN 61010 standards for safe use on high-voltage PV arrays and combiner boxes.
- ⚡ Smart Data Management Features data hold + backlit LCD for reading values in dark environments (e.g., rooftops)
- ✅ Engineered for Solar Professionals Auto-ranging simplifies operation for technicians, while IP54 dust/water resistance and low-power auto-off ensure reliability in outdoor installations.
The thermal dataset behind the results
The paper evaluates on a dataset it credits to Pierdicca et al. as the Photovoltaic Thermal Images dataset. The table lists what the paper says about it.
| Attribute | As described in the paper |
|---|---|
| Images | 1,009, each at 512 × 640 resolution |
| Labels | One expert-annotated binary mask per image |
| Site | 66 MW ground-mounted facility in Tombourke, South Africa |
| Capture window | 21–27 January 2019 |
| Conditions | Clear skies, high irradiance |
| Sensor | UAV-mounted radiometric thermal camera; make and model not named |
| Recorded thermal values | 2.25 to 103.34 °C |
Every image therefore comes from one plant during one week of Southern Hemisphere summer. The temperature range describes the values present in this dataset. It is not an operating envelope for PV modules, and it is not a camera specification.
Rank #2
- ⚡ Professional-Grade PV Testing Measures maximum power (Pmax) up to 2000W, open-circuit voltage (Voc: 12-150V), and short-circuit current (Isc: 35A) with ±0.8% accuracy, ideal for validating solar panel performance in R&D, manufacturing, and field maintenance. shading issues, helping installers maximize energy harvest and ROI for residential/commercial systems.
- ⚡ MPPT Efficiency Optimization Tracks Vmp 150V) & Amp (35A) in real-time to identify panel degradation or
- ⚡ Industrial Safety & Durability Rated CAT III 1000V/CAT IV 600V with double-insulated probes, meeting IEC/EN 61010 standards for safe use on high-voltage PV arrays and combiner boxes.
- ⚡ Smart Data Management Features data hold + backlit LCD for reading values in dark environments (e.g., rooftops)
- ✅ Engineered for Solar Professionals Auto-ranging simplifies operation for technicians, while IP54 dust/water resistance and low-power auto-off ensure reliability in outdoor installations.
Training and evaluation setup
- Optimizer and schedule: Adam, learning rate 1 × 10⁻⁴, batch size 4, trained for 200 epochs.
- Loss: an equal 0.5/0.5 combination of binary cross-entropy with logits and Dice loss.
- Augmentation: horizontal flips, brightness and contrast changes, and shift, scale and rotation transforms.
- Validation and testing: run without augmentation.
- Stability: the paper reports five repeated training runs for stability analysis.
Reported results
Every figure in this section is the authors’ measurement on this dataset and protocol. The abstract summarizes the outcome in one sentence: “The results highlight the efficacy of lightweight encoder–decoder frameworks that incorporate atrous spatial pyramid pooling, enabling accurate and computationally efficient fault detection in photovoltaic installations.” That is the paper’s own statement, not a quotation from a named author.
Free tools Windows power users keep installed
One-click scans. No signup required.
Headline metrics
| Metric | MFPN-ASPP | What it measures |
|---|---|---|
| Dice | 0.8440 | Overlap between predicted and annotated anomaly pixels |
| IoU | 0.7564 | Intersection over union of predicted and annotated anomaly pixels; usually reads lower than Dice for the same prediction |
| Precision | 0.8771 | Share of predicted anomaly pixels that are annotated anomalies |
| Recall | 0.8428 | Share of annotated anomaly pixels the model flags |
| Boundary F-score (2-pixel tolerance) | 0.8436 | Agreement between predicted and annotated edges, counting an edge as matched if it falls within 2 pixels |
Test-set results
In its per-image results, the paper reports a 101-image test set with mean Dice 0.844 ± 0.165 and mean IoU 0.757 ± 0.187. The ± values are as printed in the paper. Per-image scores vary considerably around these means, so an individual image can score well above or below the average.
Rank #3
- 【Accurate Solar Panel Tester with MPPT】 Get precise solar readings every time. This solar panel tester measures Pmax, Voc (3–90V), and Isc (0–40A) with high accuracy. Built-in MPPT tracking ensures reliable performance data for solar panel testing, troubleshooting, and system optimization.
- 【Plug & Play – Easy for Beginners】 No complicated setup. Simply connect the MC4 cables and start testing instantly. Supports both auto and manual modes, making this solar tester perfect for beginners and professionals.
- 【Large Sunlight-Readable LCD Display】 3.2-inch LCD screen clearly shows voltage, current, and power at once. Designed for outdoor use, the display remains easy to read even under direct sunlight.
- 【All-in-One Solar Power Meter & PV Tester】 Works as a solar power meter, PV tester, and photovoltaic multimeter. Ideal for solar panel installation, maintenance, fault diagnosis, and performance testing in RV, home, and off-grid systems.
- 【Multi-Protection & Durable Design】 Built-in protection against over-voltage, over-current, reverse polarity, and overheating. Shock-resistant housing ensures durability for long-term outdoor and field use.
What the ablation shows
| Configuration | Dice | IoU |
|---|---|---|
| MobileNetV2-FPN (no ASPP) | 0.6872 | 0.5234 |
| MobileNetV2-FPN + ASPP (MFPN-ASPP) | 0.8440 | 0.7564 |
The gap between these two rows is the gain the paper attributes to adding ASPP, measured on the same dataset.
Comparison with other architectures
The paper reports that MFPN-ASPP scores higher than U-Net, LinkNet, FPN and Mask-RCNN on Dice and IoU. The individual baseline values are in the paper’s comparison results, and they should be read alongside the same training and evaluation protocol the authors used for MFPN-ASPP. These are within-paper comparisons, not a ranking of these architectures across datasets or implementations.
Rank #4
- Enhanced Testing Power: Measures solar panel power output from 5W to 1800W, voltage from 20V to 120V, and current from 0A to 60A, making it suitable for a wide range of solar applications, from residential systems to commercial installations. This comprehensive range allows for quick diagnostics and optimizations, ensuring better performance and higher returns on investment.
- Doubled Voltage Range: The upgraded EY-1800W features a voltage range of 20-120V, enabling the tester to support solar panel combinations with series connections below this limit, greatly enhancing testing capabilities for complex photovoltaic systems and meeting higher voltage application scenarios.
- Ultra Clear LCD Display: Features a large, easy-to-read LCD screen that provides clear visibility of measurements even in bright sunlight, ensuring accurate data viewing during testing.
- Safety Multi-Protection: Features over-voltage, over-temperature, over-current, and reverse polarity protection, ensuring a safe testing process. No additional power supply is required, simplifying usage.
- Portable Design: Weighing only 470 grams and equipped with EVA packaging and various connecting cables, it’s easy to carry and store, and suitable for a wide range of solar applications.
Efficiency and drone inspection: what is claimed and what is shown
The authors describe MFPN-ASPP as computationally efficient and suitable for UAV inspection. That is their interpretation of a benchmark experiment, not a field trial. The paper does not provide a validated hardware benchmark that supports a specific real-time speed or energy figure for a given drone or onboard computer, so the suitability claim should be read as the authors’ view.
Quick Recap
Best Value
- MPPT Power Measurement: Designed for photovoltaic modules to accurately measure Pmax, Vmp, Imp, Voc and Isc, enabling fast and reliable performance checks during installation and maintenance.
- 2000W / 150V DC MAX: Supports up to 1000W output power, 80V DC maximum voltage and 0–45A current, suitable for most residential and commercial PV panels.
- AUTO & MANUAL Mode: Automatic mode for continuous measurements and manual mode for controlled single tests, ideal for commissioning, servicing and inspection tasks.
- 3.2" LCD Display With Backlight: Large 3.2-inch LCD screen with backlight and data hold function ensures clear readability when working outdoors.
- Portable & Ready To Use: Compact, lightweight tester with rear hanging slot. Includes test leads, alligator clips, storage pouch and user manual for immediate operation.
How to read the metrics for PV inspection
- False alarms. A precision of 0.8771 means that roughly 12% of predicted anomaly pixels fall outside the annotated anomalies. In a maintenance workflow, those are healthy areas that would still cost a site visit or a closer look.
- Missed anomalies. A recall of 0.8428 means that roughly 16% of annotated anomaly pixels are not flagged. Those are defects the model would leave unreported.
- Area versus edges. Dice and IoU reward overlapping area. The boundary F-score asks whether predicted edges land close to annotated ones. A 2-pixel tolerance corresponds to a different physical distance on each flight, depending on altitude and sensor resolution.
- Masks, not panel verdicts. These metrics score agreement between masks. They do not, by themselves, establish whether each panel is correctly classified as faulty or healthy.
Getting the data and code
- Open the article page and go to its data availability statement.
- Submit the dataset request form that the statement describes, giving your research purpose. Access is granted on that basis; the dataset is not presented as an unrestricted public download.
- For the code, contact the corresponding author named on the article. The paper says code is available upon request.
- Check the statement again before you rely on it, because access arrangements can change after publication.
Using the result responsibly
- When comparing with another segmentation model, use the same image set, annotation masks and train/test split, with matching preprocessing, loss, augmentation and metric definitions.
- Use the same 0.5 probability threshold when converting outputs to masks, or report the threshold you used.
- Test the model on imagery from your own sites, sensors and flight conditions before relying on it. The paper’s results do not transfer automatically to other installations.
- If you plan to acquire imagery, verify radiometric capability, UAV compatibility and current availability for any camera independently. The paper names no make or model, so it does not establish that a particular camera will reproduce its imagery.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
