The recent progress in recognizing low-resolution instantaneous high-density surface electromyography (HD-sEMG) images opens up new avenues for the development of more fluid and natural muscle-computer interfaces. However, the existing approaches employed a very large deep convolutional neural network (ConvNet) architecture and complex training schemes for HD-sEMG image recognition, which requires learning of >5.63 million(M) training parameters only during fine-tuning and pre-trained on a very large-scale labeled HD-sEMG training dataset, as a result, it makes high-end resource-bounded and computationally expensive. To overcome this problem, we propose S-ConvNet models, a simple yet efficient framework for learning instantaneous HD-sEMG images from scratch using random-initialization. Without using any pre-trained models, our proposed S-ConvNet demonstrate very competitive recognition accuracy to the more complex state of the art, while reducing learning parameters to only ≈ 2M and using ≈ 12 × smaller dataset. The experimental results proved that the proposed S-ConvNet is highly effective for learning discriminative features for instantaneous HD-sEMG image recognition, especially in the data and high-end resource-constrained scenarios.