Abstract: This study proposes an intelligent waste classification system based on an SE-enhanced ResNet deep neural network to achieve accurate and efficient domestic waste sorting. The proposed model integrates channel-wise attention mechanisms to improve classification performance. The structural dimensions of sorting mechanism are optimized using the closed-loop vector method, while the output power of the motor is calculated based on the principle of virtual work, providing a theoretical basis for motor selection. A channel attention mechanism is integrated into the network model to facilitate effective interaction and selection among channels, thereby enhancing the utilization of channel information. The control system employs an Arduino Mega 2560 for sensor data acquisition and communicates with an NVIDIA Jetson NX via serial interface. Combined with a throwing motion control algorithm, this enables precise control of motors. Additionally, the system features a custom-designed compression unit that mechanically compresses compressible waste to improve space utilization. To evaluate the classification accuracy, the system was tested using twelve common types of household waste. Experimental results demonstrate that the SE-ResNet-based deep neural network achieves an average recognition accuracy of up to 95.83%, meeting practical application requirements.