Plant disease identification at the leaf stage is a well-bounded problem with outsized consequences for smallholder agriculture. We train a convolutional neural network on the PlantVillage dataset — 55,306 images across 39 classes — using a custom architecture in Keras, augmented with standard geometric transforms. The resulting model reaches 98.7% validation accuracy after 40 epochs, with the confusion matrix concentrated along the diagonal and remaining errors clustered around visually proximate classes. We document the full pipeline — preprocessing, augmentation strategy, architecture, and hyperparameters — so that the work is reproducible on a commodity GPU.

Soil is the most consequential variable in small-scale cultivation, and the one smallholder farmers have the least instrumentation for. We describe an IoT-based soil-quality measurement and smart-agriculture analysis system that samples soil parameters — temperature, humidity, moisture, nitrogen, and rainfall — through a cluster of low-cost sensors wired to an Arduino Nano and a Node MCU ESP8266, with data relayed upstream for analysis. The system provides an instant read of soil-element values to support the central decision in cultivation: whether a given parcel is eligible for a given crop this season. A smart monitoring system can produce that read; the instrument array here demonstrates it using a DHT11 sensor, a YL-69 soil-moisture probe, a 5V relay, and a rainfall-humidity sensor. Collected datasets are used to analyse soil fertility, surface crop weaknesses, and recommend which plants suit a given soil type. The contribution is the parameter set chosen for multi-sensor analysis and the practical arrangement that makes it deployable in the field.