EcoSort Smart Station

What it does

Problem: Traditional garbage classification relies on manual judgment, with high error rate, low efficiency, and lack of user participation motivation, which makes it difficult to improve the recycling rate. Solution: Accurate classification Through AI image

Your inspiration

Social pain point drive Inefficient traditional classification: Although garbage classification policies are widely implemented, they rely on manual memory or simple labels, and wrong disposal is still common, resulting in pollution of recyclables and increased processing costs. Lack of motivation for continuous participation: Residents are not motivated enough because of the cumbersome classification or "no immediate return", and policy implementation tends to become a formality. 2. Opportunities for technological change The maturity of AI and the Internet of Things: Computer vision (such as ResNet,

How it works

1. Core function: Help you sort garbage like a "smart assistant" Automatically identify garbage types: When you approach a trash can, the built-in camera will take a picture of the garbage (similar to mobile phone face unlocking). The AI ​​system quickly determines whether it is plastic, paper, kitchen waste or hazardous waste (such as batteries) by comparing the "gallery" of thousands of types of garbage. After identification, the screen or voice will prompt you: "This is a recyclable plastic bottle, please put it in the blue box." Error-proofing design: If the delivery is wrong (such as throwing a plastic bag into the kitchen waste bin), the trash can will automatically alarm and remind you to reclassify. Some models support automatic sorting: the robotic arm will push the wrong items to the correct box (similar to the express sorting robot). 2. User incentives: Sorting can "make money" Points reward: After each correct sorting,

Design process

Phase 1: Concept Birth (Pain Point Analysis → Preliminary Conception) Source of Inspiration: It was observed that when there was no supervision at the community garbage sorting booth, residents often mixed up the garbage, causing the cleaners to sort it twice. Japan already has intelligent recycling machines, but the cost is high (about 50,000 yuan/unit), which is difficult to popularize. Phase 2: Prototype Development (Technical Verification → Functional Testing) 1. First Generation Prototype (PVC Board + Raspberry Pi) Materials: PVC board assembled box, Raspberry Pi 4B + 5 million pixel camera. Functional Verification: Training a simple image classification model with Python + OpenCV (accuracy is only 65%). Phase 3: Mass production improvement (user scenario adaptation → cost control) 1. Problems exposed in community pilot Inlet design: Initially it was a circular inlet, but large garbage (such as express boxes) was easy to get stuck → changed to a sliding cover inlet, supporting 20cm×30cm items. Phase 4: Modern iteration (data-driven upgrade) Dynamic recognition optimization: Collecting delivery data through the cloud, it was found that **"mask packaging"** was often misjudged as aluminum foil → Targeted training data for this category was added.

How it is different

Most rely on touch screens or complex buttons, which are difficult for the elderly/children to use. Our solution: "Non-sensing recognition": Just get close to the trash can, and AI will automatically determine the type of trash through the camera (without clicking any buttons). Voice priority: Use spoken prompts (such as "Please throw the milk tea cup into 'other trash'"), which is more intuitive than pure icons. Imported AI chips (such as NVIDIA Jetson) result in a single machine cost of more than 30,000 yuan, which is difficult to deploy on a large scale. Our solution: Using domestic AI chips (Horizon Sunrise X3), the cost is reduced by 60%, while maintaining a recognition accuracy of 90%+. Modular design: Flexible functions can be added (such as weighing modules, compression modules), adapted to different scenarios such as communities and shopping malls.

Future plans

1. Technology iteration: smarter and more user-friendly AI 3.0 upgrade Multimodal interaction: supports gesture control (wave to open the lid) and AR projection (display classification instructions directly on the ground). 2. Scenario expansion: from community to full chain Home version mini classifier: Small device on the kitchen countertop, automatically scans and compresses garbage, and synchronizes with the community station data ("You can earn points by classifying at home"). 3. Deepening of carbon ecology: making environmental protection "cash in" Financialization of carbon credits

Awards