This paper investigates a decentralized adaptive control strategy for a class of interconnected unknown non-linear systems. The idea of the strategy is based on the feedback linearizing control and perturbation estimation. A high gain observer is designed in association with each sub-system to estimate the states and a fictitious state which is defined to represent the system perturbation including the combined effect of system non-linearities, unknown system dynamics, disturbances and interactions between sub-systems. Subject to the availability of sub-system states, two local controllers - decentralized non-linear adaptive state-feedback controller (DNASFC) and decentralized non-linear adaptive output-feedback controller (DNAOFC) - are developed using a high gain perturbation observer (HGPO) or a high gain state and perturbation observer (HGSPO) respectively. The stability and error analysis of the high gain observers and the closed-loop control systems are addressed in detail. The two proposed controllers are evaluated on an interconnected non-linear system involving two inverted pendulums on carts without velocity measurements.