Application Story | BRAV-7123 Powers High-Speed Carrot Sorting System

In the agricultural product processing industry, carrot sorting has always been a major challenge—it requires speed, numerous standards, high precision, and high labor costs. Effective inspection by the human eye is virtually impossible, especially when conveyor belts move at a rate of 120 carrots per second.

Powered by the robust AI computing capabilities of the NVIDIA Jetson Orin NX, the BRAV-7123 high-speed carrot sorting system has been successfully adopted by overseas clients. By utilizing a fully automated AI vision solution to streamline the entire sorting process, it effectively overcomes the persistent drawbacks associated with traditional manual sorting.

Four Core Pain Points in the Carrot Sorting Industry

1️⃣ High‑speed production line: beyond human capability

Up to 120 carrots rush past on the conveyor belt every second, leaving no time for human operators to observe, judge and log defects. Manual inspection suffers from low efficiency. More critically, grading standards are inconsistent. Deviations occur for the same inspector between rested and fatigued states, as well as across different operators.

2️⃣ No guaranteed consistency in quality control

Carrot grading covers multiple dimensions including size, shape and color. Reliant on subjective human experience, manual operation fails to deliver quantifiable and repeatable standardized workflows. This directly causes fluctuations in finished‑product quality, hurting brand reputation and market bargaining power.

3️⃣ Delayed decision‑making and lack of data

Under the traditional model, production data is compiled manually after the fact. Management cannot gain real‑time visibility into critical KPIs such as output, pass rate and grade distribution. On‑site operational decisions are made based purely on experience rather than data‑driven insights.

4️⃣ Near‑impossible detection of fungal infection

Early-stage fungal infections show extremely subtle signs, such as mold spots as tiny as 0.3 mm. Human eyes can hardly identify such defects on high-speed lines. Once contaminated products reach the market, they pose food-safety risks and substantial economic losses.

Solution: How does the BRAV‑7123 deliver disruptive performance?

🧠 Core Hardware: Empowered by Jetson Orin NX, Surge in Computing Performance

Powered by the NVIDIA Jetson Orin NX module, the BRAV‑7123 delivers 70‑157 TOPS of AI computing performance, equipped with 8 GB / 16 GB on‑board LPDDR5 memory and up to 102.4 GB/s bandwidth. The conveyor‑belt encoder feeds real‑time belt‑speed and position signals to BRAV‑7123 via the CAN FD interface. As carrots arrive at predefined capture positions, the system triggers 1‑2 channels of POE industrial cameras for synchronous image capturing over Gigabit Ethernet. Multi‑angle imaging (top‑view plus side‑view) fully captures each carrot’s length, width, color and fine surface details.

Captured image data is directly fed into the Jetson Orin NX module onboard BRAV‑7123. Leveraging its 1024‑core NVIDIA Ampere‑architecture GPU and 32 Tensor Cores, deep‑learning‑based object detection and grading models run to complete the following tasks within milliseconds: carrot object detection & counting | dimension measurement | appearance grading | fungal infection identification.

Full‑interface compatibility for industrial production lines: end‑to‑end synchronous acquisition, sorting and data transmission

• Gigabit Ethernet ports connect PoE industrial cameras mounted above and alongside the conveyor belt, delivering both power supply and high‑speed image data transmission.
• CAN FD interface links to the conveyor‑belt encoder to acquire real‑time belt speed and position signals, enabling precise synchronization between image capture and carrot positioning.
• RS485 interface connects to production‑line PLCs or grading actuators. It converts AI grading results into control commands to drive subsequent sorting operations.
• HDMI interface connects to on‑site large‑screen displays for real‑time visualization of production statistics and dashboards.
• M.2 B‑key slot supports 4G/5G modules for cloud upload of production data and remote operation & maintenance. Meanwhile, the M.2 E‑key slot accommodates Wi‑Fi modules for local wireless network access, facilitating on‑site commissioning and data connectivity.
• Up to 6 × USB3.0 ports support expansion for additional industrial cameras as well as debugging and maintenance devices.

Software Layer

1️⃣ Improved YOLOv8 + DeepSORT Tracking Algorithm

Enables continuous detection of carrots on moving conveyor belts. Detection outputs are quantitatively processed into precise numerical data.

2️⃣ Fine‑grained Image Classification Model (based on YOLOv8 detection bounding boxes)

Specially trained to identify early‑stage fungal infection features such as mold spots and discoloration, detecting defects as small as 0.3 mm.

3️⃣ Multi‑task YOLO Architecture — YL‑Net

Delivers millimeter‑accurate prediction of carrot geometric properties (length, width) under motion, with a processing speed of over 80 FPS.

4️⃣ NVIDIA DeepStream for Multi‑stream AI Pipeline Construction

Efficiently processes visual data from multiple PoE cameras. It supports parallel decoding and preprocessing of multiple video streams, zero‑copy data transfer for reduced latency, as well as hardware‑accelerated inference and post‑processing.

Solution Value: Cost reduction, quality improvement, risk mitigation, and digitalization—all achieved in one step.

✅ Full workforce liberation and drastic labor-cost reduction

Traditional workflows rely on large numbers of workers for individual inspection and sorting on production lines. After deploying this system, the full inspection‑and‑sorting process becomes fully automated, shifting from human visual inspection to machine vision. Labor costs are cut by more than 80%. For large‑scale producers contributing 40% of regional output, this translates to annual labor‑cost savings in the millions.

✅ Standardized quality control for higher product premium

The AI model applies fully consistent judgment criteria free from human variables such as fatigue, mood or varying operator experience. Every carrot is graded against unified standards for stable and controllable finished-product quality. This helps build brand credibility and unlock higher market premiums.

✅ End‑to‑end real‑time data visualization, ending experience‑based production management

On‑site managers gain instant visibility into key metrics including output, grade distribution and infection rate via large‑screen dashboards, enabling fast adjustment decisions. Every insight — from dropping pass rates for certain batches to production lines requiring maintenance shutdowns — is data‑backed.

✅ Greatly mitigated food‑safety risks

The BRAV-7123 vision system performs real-time, full-coverage automatic screening for early-stage fungal infections in carrots, historically hard to control. Defective carrots with tiny mold spots are automatically rejected during sorting. This prevents faulty goods from reaching end markets, and proactively avoids multiple operational risks including food-safety incidents, customer claims and negative brand publicity.

Automation and digitalization of large‑scale fruit and vegetable sorting represent an inevitable industry trend. The BRAV‑7123 High‑Speed AI Carrot Sorting System has completed on‑site validation with overseas customers.

As an edge‑AI computing platform, the BRAV‑7123 delivers end‑to‑end synergy with robust 70‑157 TOPS AI computing power, millisecond‑level real‑time inference, and comprehensive industrial interfaces. It safeguards stringent quality‑control requirements for every carrot on high‑speed production lines.

Learn more about BRAV-7123 series