AI Computer Vision System for Plant Disease Detection in Agriculture

Development of an AI System for Plant Disease Detection

Phytophthora on tomatoes under favorable weather destroys 80% of the crop in 10 days. The problem is not that the disease is incurable — the problem is that by the time an agronomist visually notices symptoms, the infection has already spread to 15–20% of the leaf surface. We develop computer vision systems that detect infection at 3–5% leaf area, providing a 3–4 day window for treatment. Our years of experience and accuracy guarantee of at least 90% allow agricultural holdings to reduce crop losses by 30–50%. The system pays for itself in less than one season due to reduced crop losses. Savings on plant protection products can reach 40%. We have already completed 15+ projects in this field. Typical project cost ranges from $10,000 to $50,000 depending on scope and complexity.

How to Improve Plant Disease Detection Accuracy?

The main technical challenge is inter-class similarity of symptoms. Iron deficiency chlorosis and early stage powdery mildew on cucumber leaves look virtually identical — both produce light spots with blurred edges. A model trained only on spot color systematically confuses them.

Why Multi-Feature Analysis Outperforms RGB Classification?

We use multi-feature analysis including:

• Textural features of the spot (LBP, Haralick features via skimage.feature)
• Lesion geometry: shape, perimeter-to-area ratio
• Spread pattern across the leaf (marginal vs central vs diffuse)
• Vegetation stage as context (young leaves give different patterns)

A dual-stream architecture works well: one stream processes an RGB patch via EfficientNet, the other processes computed texture maps via a lightweight MobileNetV3. Output embeddings are concatenated before the final classification head. On the PlantVillage dataset plus our own annotations across 14 crops, such architecture yields top-1 accuracy 0.93 vs 0.87 for single-stream EfficientNet-B4 — a 1.07x improvement.

What to Choose: Detection or Segmentation?

For field conditions, detection is needed, not just patch classification. YOLOv8 or RT-DETR return a bounding box with disease class and confidence in real time — critical for a mobile app that agronomists use directly in the field.

From practice: YOLOv8m on a dataset of 12,000 annotated leaf images (8 diseases, 4 crops) after fine-tuning showed mAP50 = 0.81. Problematic classes — early blight and alternaria (IoU < 0.5 on 23% predictions) due to overlapping lesions. Solution: instance segmentation instead of bbox — Mask R-CNN or YOLOv8-seg allows separating overlapping foci.

YOLOv8n INT8 works 3.5x faster than RT-DETR-L with comparable accuracy (0.73 vs 0.84, but on a mobile device speed is more important).

Deployment in Field Conditions

Two scenarios with different requirements:

Mobile app for agronomist. Model — ONNX + ONNX Runtime Mobile on Android/iOS. YOLOv8n INT8 runs on Snapdragon 8 Gen 2 with a latency of 45–70 ms, which is acceptable for handheld shooting. The app works offline, results sync when connected.

Drone system. NVIDIA Jetson Orin NX (16 GB) + TensorRT engine. Real-time inference at 4K 30 fps, detection on every 5th frame, coordinates of infected points written to GeoJSON.

Performance comparison on different platforms:

Data: Labeling and Dataset Expansion

Public labeled datasets on plant diseases are sufficient to start (PlantVillage — 87,000 images, 38 disease classes). However, for specific crops and regions, fine-tuning on own data is needed — diseases look different in different climatic zones and on different varieties.

Data Requirements for Model Training

To train a model, you need a representative set of images from your fields. We recommend at least 500 labeled images per disease class. We speed up labeling via active learning: after initial training, the model selects the most "uncertain" examples (entropy sampling), and the agronomist labels only those. In practice, this reduces manual labeling effort by 40–60%.

What's Included in the Project

Our turnkey solution includes:

• Data audit and disease list definition
• Field data collection and annotation (if needed)
• Model development and training (transfer learning)
• Field validation and iterative improvement
• Deployment in mobile app or on edge device (Jetson, etc.)
• Performance drift monitoring, support, and retraining
• API documentation and operational guides
• Training for agronomists on system use

How Development Works: Step-by-Step Plan

1. Data audit and disease list definition
2. Field data collection and labeling (if necessary)
3. Model development and training (transfer learning)
4. Field validation, iterations
5. Deployment in mobile app or on edge device
6. Drift monitoring, support and retraining
7. API and operation documentation
8. Training agronomists to use the system

We deliver turnkey projects: from data collection to deployment. We will evaluate your project within 2 days. Contact us for a consultation. Order a preliminary audit of your project.

Timeline and Cost

Detection system for one crop and 5–8 diseases: 6–10 weeks with labeled data available. Multi-crop platform with mobile app and server component: 3–4 months. Cost is calculated individually; typical range $10,000–$50,000. Our company has been on the market for 5 years and has completed 15+ projects in machine vision for the agricultural sector.

How Distribution Shift Kills CV Model Metrics in Industry

On a production line, a camera is installed to control product quality. The model is trained on 10,000 labeled images—test accuracy mAP 0.84. Deployed to production, and in the first week it misses 30% of defects. Lighting on the line changes between shifts; distribution shift nullifies the metrics. This is a classic story with computer vision in industry, where pattern recognition fails without proper drift handling.

Our engineers, with experience from 60+ computer vision projects, know how to eliminate such scenarios. We guarantee stable model performance under real conditions.

Object Detection: YOLO, RT-DETR, and Everything in Between

YOLO is the standard for real-time detection. YOLOv8 and YOLOv11 from Ultralytics are the most used versions in production: simple API, active community, built-in validation, and export to ONNX/TensorRT. For tasks with high accuracy requirements and less critical latency, RT-DETR, a transformer-based architecture without NMS, gives better mAP on COCO at comparable speed to YOLOv8l.

A typical mistake when training a detector: dataset of 8000 images, 3 classes, fine-tune YOLOv8m—F1 0.73 on validation. Look at confusion matrix—one class is almost never detected. Cause: imbalance 1:23. Solution: oversampling rare class, focal loss for objectness, augmentations (Mosaic, MixUp disabled for rare class as they "blur" it). Transfer learning is mandatory: pretrained on COCO weights reduces data requirement by 10 times. Fine-tuning on 500–2000 domain images yields a working model in 1–2 days on a single GPU.

For edge deployment: export to ONNX → TensorRT engine. YOLOv8n in TensorRT FP16 on Jetson AGX Orin gives 150+ FPS at P99 latency < 8 ms—3 times faster than ONNX Runtime without TensorRT. On server A10G: 700+ FPS for YOLOv8n in TensorRT INT8.

How Does Fine-Tuning YOLO Help in Pattern Recognition?

Suppose you need to find micro-defects on a metal surface—a task with high resolution and class imbalance. We use YOLOv8m pretrained on COCO and fine-tune on 2000 proprietary images. Apply augmentations Mosaic, MixUp, random perspective. After 200 epochs, mAP 0.5 reaches 0.93. Key techniques:

• Focal loss for the objectness head—reduces contribution of easily classified examples.
• Class-balanced sampling—equalizes representation of rare classes.
• Test Time Augmentation (TTA)—increases recall by 5–7% through averaging over flips and scales.

Get a consultation on architecture selection for your task—contact us.

Segmentation: SAM, Mask R-CNN, and Instance Segmentation

SAM (Segment Anything Model) from Meta changed the approach to segmentation. SAM 2 works with video, supports object tracking across frames—for interactive object selection by point or bbox, it's the best out-of-the-box choice. For production instance segmentation without interactive prompting, Mask R-CNN or YOLOv8-seg are used. YOLOv8-seg trains like a regular detector with additional masks, convenient in the same pipelines. Semantic segmentation (each pixel is a class) uses SegFormer, DeepLabV3+. SegFormer-B5 provides a good balance of accuracy and speed for satellite imagery or medical segmentation.

Case study: cell segmentation on microscopic images. Dataset of 400 images with manual annotation. Training Mask R-CNN on ResNet-50 backbone gave IoU 0.61—poor. Problem: objects (cells) overlap; standard NMS kills overlapping predictions. Solution: switch to cellpose (specialized architecture for biomedical tasks) + soft-NMS. IoU increased to 0.79.

OCR: When Tesseract Fails

Tesseract is a starting point for simple tasks: printed text, good lighting, straight layout. As soon as there are handwritten elements, non-standard fonts, perspective distortions, or multi-column layouts, Tesseract degrades quickly.

PaddleOCR is a production-grade solution: text block detection + recognition + structural analysis. Works out of the box for 80+ languages, including Russian. Supports tables and complex document structures. TrOCR (Microsoft) is a transformer OCR with strong results on handwritten text. For Russian handwritten text, fine-tuning is needed: the base model is trained mostly on Latin script.

What to Do When Tesseract Cannot Handle Pattern Recognition on Documents?

For tasks like "extract data from invoices/contracts/passports," we use LayoutLMv3 or Donut—these models understand document layout, not just text. Integration via Hugging Face Transformers, fine-tuning on 200–500 annotated documents. Typical pipeline:

1. Preprocessing: deskew, denoising, binarization via OpenCV.
2. Text block detection: PaddleOCR detection or CRAFT.
3. Recognition: PaddleOCR recognition or TrOCR.
4. Post-processing: normalization, validation via regex or LLM for structured fields.

For documents with fixed structure, template matching + OCR by coordinates is often more reliable than an end-to-end solution.

Face Recognition: Identification and Verification

Face recognition = detection + alignment + embedding + matching. Each stage matters.

Detection: RetinaFace or InsightFace for accurate face localization and keypoints. MTCNN is older but reliable. Embedding: ArcFace (InsightFace) is state-of-the-art for face recognition embeddings. Models iresnet50/iresnet100 pretrained on MS1MV3 (5M identities). Embedding vector 512 float32, comparison by cosine similarity. Threshold tuning: decision threshold is a critical parameter. At threshold 0.6, typical FPR on LFW benchmark is 0.001, TPR is 0.985. In production, threshold must be calibrated to the real distribution: people in masks, with changed appearance, different lighting conditions. Liveness detection is mandatory: MiniFASNet—lightweight model on CPU; FaceX-Zoo contains several pretrained liveness detectors.

Video Analytics

Video is a sequence of frames plus a temporal dimension. A naive approach—detecting on every frame—is expensive.

Tracking: ByteTrack and BoT-SORT are the standard for multi-object tracking. They work on top of any detector, adding persistent IDs to objects across frames—enabling object counting, motion tracking, velocity.

Optimization: not every frame needs processing. For static scenes, detect every 5–10 frames, with tracking in between. For event detection (person entering a zone), background subtraction (OpenCV MOG2) serves as a lightweight pre-filter before neural detection. Action recognition: SlowFast, VideoMAE for action classification. Heavy models—for production use ONNX export + TensorRT or offline processing.

How to Measure Pattern Recognition Model Quality in Production?

Quality monitoring is key to MLOps. We track:

• Prediction confidence distribution.
• Share of low-confidence predictions (indicator of OOD data).
• Drift of input images via feature distribution (embeddings from backbone).

A drop in average confidence from 0.87 to 0.71 over a week is an early signal of distribution shift. NVIDIA Triton Inference Server recommends tracking these metrics via Prometheus. Our certified engineers set up monitoring and guarantee SLA for inference quality.

Deployment of CV Models

For online inference, we use Triton Inference Server (NVIDIA)—production standard for serving CV models. Supports TensorRT, ONNX, PyTorch, dynamic batching, multiple instances. REST and gRPC API. We guarantee stable operation under load.

Edge deployment: ONNX Runtime on ARM/x86 CPU. TensorFlow Lite for mobile devices. OpenVINO for Intel CPU/GPU/VPU—gives 2–3× speedup on Intel hardware compared to ONNX Runtime. After deployment, we hand over the model with documentation and train personnel.

What Is Included in the Work

Deadlines and Cost

A prototype detector on existing data takes 1–2 weeks. Production system with optimization for target hardware takes 4–8 weeks. Full cycle including data labeling (1000–5000 images) takes 2–4 months. Cost is calculated individually for each task. Typical savings from implementing a quality control system can be significant per production line.

We have been in the market for over 5 years and completed 60+ computer vision projects. We will evaluate your project end-to-end—request a consultation to get a quote and technical proposal.