This podcast episode will explore the YOLO_MRC model, a deep learning model that can detect and count pests in real-time using images. The model was developed to address issues with existing pest detection methods, such as:

🍈Long inference times: The time it takes for the model to process an image and make a prediction.

🍈 Low accuracy: The ability of the model to correctly identify pests.

🍈 Large model sizes: The amount of storage space the model requires.

The YOLO_MRC model is based on the YOLOv8n model and includes three key improvements:

👉🏼 Multicat Module: This module helps the model focus on the target by incorporating an attention mechanism.

👉🏼 Reducing Detection Heads: The number of detection heads in the model is reduced from three to two, decreasing the number of parameters.

👉🏼 C2flite Module: This module enhances the model's ability to extract deep features.

These modifications enable YOLO_MRC to achieve faster processing times, higher accuracy, and a smaller model size compared to the original YOLOv8n model.

Testing and Results

The researchers tested YOLO_MRC on a dataset of Bactrocera cucurbitae pests, which affect melon, fruit, and vegetable crops. The dataset consisted of images captured from videos of trap bottles. The model was compared to four other detection models:

● YOLOv5s-ECA

● Fast-RCNN (Mobilenetv2)

● YOLOv5Ghost

● YOLOv7Tiny

YOLO_MRC achieved the best performance in terms of processing time, recall, and model size. It also had the highest accuracy when compared to manual counting results, with an average accuracy of 94%.

Real-time pest detection and counting can benefit agriculture in several ways:

↳ Early pest detection: Enables timely intervention and prevents widespread infestations.

↳ Optimised pesticide use: Reduces pesticide waste and environmental pollution by providing accurate pest counts.

↳ Data for pest management: Provides valuable information for agricultural managers to make informed decisions.

Limitations and Future Research

The YOLO_MRC model has some limitations:

● It is currently only applicable to Bactrocera cucurbitae pests.

● It may not be accurate in all outdoor environments.

● It can have errors in cases of overlapping occlusions.

Researchers plan to address these limitations in future research by:

🟡 Improving the model's accuracy for multi-class pest detection.

🟡 Optimising the model's adaptability to different environments.

🟡 Enhancing the model to handle overlapping occlusions.

🟡 Exploring applications on mobile devices for use in the field.

Conclusion

The YOLO_MRC model offers a promising solution for real-time pest detection and counting. Its compact size, high accuracy, and fast processing speed make it suitable for practical use in agriculture. Further research and development will enhance its capabilities and expand its applications.

Hosted on Acast. See acast.com/privacy for more information.

In this episode, we dive into the cutting-edge world of smart farming and explore how artificial intelligence (AI) is transforming agriculture, specifically focusing on pitaya (dragon fruit) farming in China.

Our discussion centers around a recent breakthrough in precision agriculture with the development of GSE-YOLO, a lightweight, high-precision model based on YOLOv8n. This innovative AI technology is designed to detect the ripeness of pitaya in natural environments, significantly improving efficiency and reducing labor costs.

We break down the key advancements in the GSE-YOLO model, including the integration of GhostConv, SPPELAN, and EMA attention mechanisms that make the system more robust and faster, while maintaining high accuracy. Learn how this technology addresses challenges in fruit detection, such as varying lighting conditions, overlapping fruits, and the natural complexity of the farm environment. With an 85.2% detection accuracy and an mAP50 of 90.9%, GSE-YOLO is a game changer for the future of farming.

Tune in to discover how these technological advancements could lead to more sustainable farming practices, reduce waste, and potentially revolutionize how we harvest crops across the world.

Hosted on Acast. See acast.com/privacy for more information.

In this episode, we dive into how advanced weather forecasting models are transforming precision agriculture. From improving water use to boosting crop yields, cutting-edge models like Temporal Convolutional Neural Networks (TCNN) and CatBoost are helping farmers make smarter decisions.

🔍 Key takeaways:

• 140,160 agrometeorological data points from Morocco's Sidi Rahal region were used to build forecasting models based on air temperature, solar radiation, and relative humidity.
• The TCNN model achieved remarkable accuracy, with an RMSE of 0.88°C for 1-day air temperature forecasts, outperforming traditional models.
• For solar radiation, TCNN achieved an RMSE of 29.26 W/m², allowing farmers to fine-tune irrigation schedules and conserve water.
• CatBoost excelled in longer forecasts, with an RMSE of 25.20 W/m² for 1-week solar radiation predictions, showing how machine learning can improve farming efficiency.

🌦 Why it matters:

Farmers can now plan irrigation with 1-day to 3-day precision, saving water and reducing costs in regions with limited resources.

Accurate forecasts prevent crop losses and boost yields by responding quickly to changing weather conditions.

💡 What you'll learn:

1. How data-driven models like TCNN and CatBoost are shaping the future of agriculture.
2. The critical role of weather forecasting in improving resource management and sustainability on farms.
3. The impact of AI and machine learning on adapting agriculture to climate change and resource scarcity.
4. Join us to explore the future of farming, where weather tech meets AI innovation! 🎙️🌍

🔔 Subscribe now to stay updated on how AI is revolutionizing agriculture!

Any questions or ideas?

Drop us an email to support@petioleapp.com

Install Petiole Pro:

https://play.google.com/store/apps/details?id=com.petiolepro.farmvision&hl=en_GB&gl=US

Find more about Petiole: https://petiolepro.com

Get calibrating plate: https://petiole.ai

Our Linkedin: https://www.linkedin.com/company/petiole

Our Researchgate: https://www.researchgate.net/lab/Petiole-Pro-Lab-Maryna-Kuzmenko

Our Instagram: https://www.instagram.com/petiolepro/

Our Facebook: https://facebook.com/petiole.co

Our Twitter: https://twitter.com/petioleapp

Our TikTok: https://tiktok.com/@petiolepro

#petiolepro #aiinagriculture #weatherforecast

Hosted on Acast. See acast.com/privacy for more information.