This paper proposes a lightweight convolutional neural network (CNN) model for the classification of four types of cassava leaf diseases: cassava mosaic disease (CMD), cassava green mottle (CGM), cassava bacterial blight (CBB), and cassava brown streak disease (CBSD), along with the healthy class. The model uses depthwise separable convolution to reduce the number of parameters, making it suitable for mobile devices. It incorporates both channel attention and spatial attention mechanisms to highlight diseased areas and suppress background noise. The model also uses inverted residual blocks to further reduce parameters and improve efficiency. Experimental results show that the proposed model outperforms state-of-the-art models like VGG16, ResNet 50, EfficientNet, MobileNet V1, and MobileNetV2, using 1.1 million fewer parameters than MobileNet V2. The model uses natural scene images for training and testing, achieving a classification accuracy of 77% on a dedicated test dataset. The model is designed for deployment on low-computing-power devices, such as smartphones, to assist farmers in detecting and classifying cassava leaf diseases efficiently. The paper also discusses the importance of cassava as a staple food in Africa and other regions, and the impact of diseases on cassava production. The model's contributions include the development of a lightweight CNN model for cassava leaf disease classification, the use of attention mechanisms to enhance performance, and the use of natural background images for training. The model's effectiveness is demonstrated through experimental results, showing its potential for real-world applications in agriculture.This paper proposes a lightweight convolutional neural network (CNN) model for the classification of four types of cassava leaf diseases: cassava mosaic disease (CMD), cassava green mottle (CGM), cassava bacterial blight (CBB), and cassava brown streak disease (CBSD), along with the healthy class. The model uses depthwise separable convolution to reduce the number of parameters, making it suitable for mobile devices. It incorporates both channel attention and spatial attention mechanisms to highlight diseased areas and suppress background noise. The model also uses inverted residual blocks to further reduce parameters and improve efficiency. Experimental results show that the proposed model outperforms state-of-the-art models like VGG16, ResNet 50, EfficientNet, MobileNet V1, and MobileNetV2, using 1.1 million fewer parameters than MobileNet V2. The model uses natural scene images for training and testing, achieving a classification accuracy of 77% on a dedicated test dataset. The model is designed for deployment on low-computing-power devices, such as smartphones, to assist farmers in detecting and classifying cassava leaf diseases efficiently. The paper also discusses the importance of cassava as a staple food in Africa and other regions, and the impact of diseases on cassava production. The model's contributions include the development of a lightweight CNN model for cassava leaf disease classification, the use of attention mechanisms to enhance performance, and the use of natural background images for training. The model's effectiveness is demonstrated through experimental results, showing its potential for real-world applications in agriculture.