Уровень 0 · материалов: 5
В кластер входят документы о технической реализации сложных систем на базе микроконтроллеров или бюджетного оборудования с акцентом на оптимизацию, и не входят документы о конфиденциальности данных, пазлах, конкурсах или методах подавления помех в антеннах.
Общие признаки: реализация сложных функций на микроконтроллерах, минимизация аппаратных требований, оптимизация ресурсов памяти и вычислений, выполнение задач ИИ на маломощном железе
Группа выше: Оптимизация производительности кода
Смысл: The core idea is to prove that a functional GPS receiver, including raw signal processing and coordinate calculation, can be implemented on general-purpose microcontrollers (STM32/ESP32) without specialized ASIC hardware, provided that computationally expensive operations are optimized using bitwise logic.
The author successfully built a DIY SDR GPS receiver using STM32 and ESP32 microcontrollers by optimizing signal correlation with XOR operations and implementing a multiplexed tracking system.
Смысл: The core idea is to demonstrate the absolute minimum hardware requirements for neural network inference by successfully implementing a digit recognizer on a nearly primitive 8-bit microcontroller. It highlights the trade-off between model precision, memory constraints, and the necessity of low-level assembly optimization.
A researcher successfully implemented an MNIST digit recognition neural network on a 3-cent Padauk microcontroller by using 8x8 downscaling, 2-bit quantization, and custom assembly code.
Смысл: The text describes a technical method for implementing True RMS voltage and frequency measurement using an 8-bit microcontroller. It emphasizes the difference between average and RMS values and provides the mathematical and hardware logic needed for an accurate implementation.
A technical guide on implementing True RMS voltage and frequency measurement using an 8-bit MCU and a differential amplifier, emphasizing accuracy over simple average methods.
Смысл: The main idea is to demonstrate that high-performance AI tasks like object detection (YOLO) can be executed on extremely small, low-cost hardware (Luckfox Pico Mini) by leveraging a dedicated NPU and optimizing software/memory usage.
A detailed technical guide on deploying custom Yolov8 object detection on the Luckfox Pico Mini using its NPU, including hardware setup, C++ programming, and model quantization.
Смысл: The author describes the creation of a simple machine learning system that monitors public webcams to notify users when the bridges in St. Petersburg are closed, bypassing unreliable official schedules. The main idea is that a minimal approach (logistic regression and basic image processing) can be highly effective for solving specific, practical problems.
The author created a practical bridge-monitoring system using public webcams and a simple logistic regression model to notify users of bridge closures in St. Petersburg.