High-performing image classification models have unaffordable computation and energy costs for resource-limited platforms. As a solution, reservoir computing based on cellular automata has been proposed. This research improves the previous work with enhancements at both the algorithmic and architectural level. Using a random forest classifier with binary features reduces computation to a fraction. We reduce memory usage by identifying and pruning the least important features. An architecture with an increased level of parallelism which processes images in a single pass reduces memory accesses and FPGA logic. These improvements have a minor accuracy tradeoff.