Recognition of boiling sheep based on Seeed XIAO ESP32

The winter vacation task I choose is to make a smart home based on Seeed XIAO ESP32 to detect pet activities.、

First of all, the Seeed XIAO ESP32 development board is introduced. It belongs to a small development board, which integrates camera sensor, digital microphone and SD card support.It is ESP32S3 32-bit, dual-core, Xtensa processor chip, running at a maximum of 240MHz, installing multiple development ports, supporting Arduino, and it has a high-resolution camera, great memory and excellent RF performance.With embedded ML computing power and photography ability, it can be an excellent tool for developing visual language and visual AI.These excellent performance and excellent product quality have laid a solid foundation for the completion of my winter vacation task.

Combined with the theme of intelligent pets in this winter vacation task, I simply designed a smart furniture that can detect the house pets sneaking into the kitchen and will have lights to warn the pets.The design idea is mainly divided into the following three blocks: 1. Train the required detection model dataset through roboflow. Because we need to detect domestic pets, we need to train our dataset with photos of domestic pets. There are two ways to obtain photos of domestic pets. The first is to grab photos of domestic pets through Seeed XIAO ESP32 combined with Arduino and store them in SD card. The specific code will be given in the design resources. The model trained by the photos obtained in this way has a better effect. The other way is to find photos of various breeds of cats on the Internet to expand our dataset, so as to ensure that there are other cats invading our kitchen and have the same alarm effect. Next, I will introduce how to give the dataset we need through roboflow. First, open the roboflow platform and create a new project. Here I use a pillow instead of a pet. This is the new project I have created.

Following that, we proceed by selecting the file, uploading the necessary image data, and clicking on "Save and Continue" at the upper right corner. We then patiently await the successful upload of the files. Once the upload is complete, we move on to select the images for annotation. After completing the annotation, we download the annotated dataset and prepare the next step of training the dataset by copying its URL.

Subsequently, we proceed to log in to the AI Studio platform in order to train our model. Within the ModelAssistant project, which is built upon the YOLOv5 framework, we meticulously train our dataset. By carefully following each step, we successfully obtain the desired target files.

Throughout this winter break task, I encountered a significant challenge in immersing myself into an unfamiliar subject and gradually gaining proficiency. However, I was fortunate to overcome this challenge with the invaluable assistance of the live tutorials conducted by the official instructors. Additionally, the comprehensive learning materials provided on the Baidu Cloud platform played a crucial role in my training journey. These resources not only enabled me to train my own model step by step but also made the entire process accessible to beginners like me. The thoughtful and meticulous design of the tutorials deserves praise.

Looking towards the future, I express my hope that the authorities will release more exceptional development boards to further enrich the winter break tasks. This would provide me with the opportunity to continue expanding my knowledge through their well-crafted tutorials.